Two CORS guides here. This is the general one: a self-hosted cors-anywhere style proxy for any URL and any method, including POST. If you just need to pull data files into the browser (GitHub, Google Drive, Dropbox, Yahoo Finance), the simpler GET-only file proxy is here: Free CORS Proxy for Yahoo Finance and Any API.
\n\nA lot of people look for a cors-anywhere alternative or replacement they can run themselves. Something general: give it any URL, any method, including a POST with a JSON body, and it adds the CORS headers and passes the call through. The well known ones are the Node project cors-anywhere and its Cloudflare port cloudflare-cors-anywhere.\n\nThis post gives you that, written my way: a small, free Cloudflare Worker that works as a general CORS proxy. You run your own copy. No server, no custom domain, free Cloudflare account is enough (100k requests per day). The full code is below, and it is also in a GitHub repo you can clone or hand to your AI coder.\n\nIf you have not read the main CORS post yet, start there: Free CORS Proxy for Yahoo Finance and Any API. It explains what CORS is in plain words, the preflight nuance, the no-cors trap, and when to use a Worker vs a Vercel serverless function vs a FastAPI backend. This post assumes you already know why you need a proxy and just want the general purpose one.\n\nHow this is different from the file proxy\n\nThe Worker in the main post is on purpose a narrow one. It was built to pull data files into a browser app (GitHub, Google Drive, Dropbox, and Yahoo Finance if you add it). So it only does GET, and it only fetches from a small allow list of file hosts.\n\nThis one is the general version. The difference is scope:\nIt passes through any method, so POST, PUT, PATCH, DELETE all work, not just GET.\nIt forwards the request body, so a JSON POST goes through.\nIt can fetch any URL (you decide how wide to open it).\n\nSame idea, bigger job. If you just need to pull files into the browser, the file proxy is simpler and safer. If you need to proxy a real API call, including POST, use the one here.\n\nHow you call it\n\nOnce it is deployed, put your target URL, URL-encoded, in the url query parameter:\n\n``\nhttps://your-worker.workers.dev/?url=\n 🎬 Try the Live Dashboard →\n Cine Pro\n
\n\nBuilt with AI assisted coding (Claude Code now). Handles drill-down, side-by-side comparisons, instant search across 15 million people, similar movie recommendations using Jaccard similarity, career timelines spanning decades.\n\nIf you're building a custom dashboard, especially with multi-GB database, the patterns here apply and components can be extracted and adapted. In Part 2 I would go deeper into query optimization - how queries taking 5-9s were reduced to 500ms to 3s.\n\nThe Backend: FastAPI + DuckDB\n\nWhy DuckDB?\n\nEmbedded database - no server process to manage. Analytical queries work out of the box and are fast.\n\nDatabase Setup\n\nDuckDB file sitting on my Hetzner VPS at /data/imdb.duckdb. I use duckdb.connect() directly - no ORM. Each query opens a connection, runs, closes. DuckDB is embedded, so connection overhead is nothing, like ~1ms.\n\nTwo Endpoint Types\nRead-only endpoint (/api/query/{filename})\nSQL sanitization strips DELETE, DROP, INSERT, UPDATE, ALTER\nReturns JSON results\nRate limited: 20/second burst, 500/minute sustained\nQuery timeout: 30 seconds\nIn-memory cache: 2000 queries per file, 30-day TTL\nAdmin endpoint (/api/admin/query/{filename})\nSeparate API key\nNo SQL restrictions\nNo timeout (allows long CREATE TABLE, bulk operations)\nUsed for building optimization tables and data pipeline tasks directly on server\n\nThe database is 16GB. Admin endpoint lets me run transformations in place - create denormalized tables, compute aggregates, build indexes - all server-side.\n\nRate Limiting\n\nPer-IP limiting using slowapi. Burst limit is 20/second - high because dashboards send 10-15 parallel queries on page load. Sustained limit is 500/minute - prevents abuse while allowing heavy usage. For IP tracking I use a custom header (X-Original-Client-IP) because Cloudflare was overwriting standard headers like X-Forwarded-For. Custom headers pass through unchanged.\n\nFrontend: React + Vercel\n\nReact app deployed on Vercel.\n\nMix of React Context and local component state. SettingsContext holds global preferences (backend selection, theme). Page-level data (search results, filmography, stats) lives in local useState - each page manages its own data.\n\nQuery Composition\n\nSQL queries are template strings built in the frontend. The component constructs the query, sends it to the serverless proxy, proxy forwards to backend, results come back as JSON arrays.\n\n``javascript\nconst query = \n SELECT primaryTitle, startYear, averageRating\n FROM person_filmography\n WHERE nconst = '${personId}'\n LIMIT 5000\n\nconst result = await executeImdbQuery(query)\n`\n\nData Flow\n\nUser interaction → Component builds SQL → Serverless proxy call → Backend executes → Results stored in component state → Render\n\nAll backend calls route through /api/duckdb serverless function. No direct database access from browser.\n\nFrontend-to-Backend: The Serverless Proxy\n\nReact frontend on Vercel. Backend URL never exposed in browser code. All API calls route through a Vercel serverless function (/api/duckdb) that acts as a proxy.\nFrontend calls /api/duckdb?action=imdb-query on its own domain\nServerless function holds the backend API key in server-side env vars\nFunction forwards request to FastAPI backend\nBackend returns results through serverless to frontend\n\nThree layers of protection:\nBackend URL hidden from browser\nBackend API key never exposed client-side\nOptional: Clerk authentication at serverless layer (JWT validation before forwarding)\n\nFor low-security apps (public demos, personal projects), the API key alone is sufficient. For high-security needs, add JWT validation at both serverless and backend layers. Full details in Part 3 of my 2026 infra guide.\n\nAuth Toggle\n\nI built Clerk authentication with an on/off switch (AUTH_ENABLED=true/false env var). When disabled, app loads ~250ms faster - no Clerk SDK bootstrap, no token verification. Useful for public demos.\n\nFrontend Components\n\nReusable for other dashboards. Download the repo, point your AI coder at a component and have it adapt it for your requirements.\n\nDashboard Overview\n\nStatic stat cards, charts, and pre-computed aggregates pulled from single JSON blob.\n\nSearch\nType-as-you-search autocomplete\nSmart Search: keeps the default search fast. IMDb has 15 million people, but most are one-time appearances. The toggle filters down to 2.3 million people who have 3 or more credits. I expect to cover 80-90% of searches. Search loads in 200ms instead of 2 seconds. If someone doesn't show up, turn off Smart Search to search everyone.\nAdvanced filters (era, genre, votes threshold) with auto-reset when typing\n\nProfile Analysis\nComplete filmography table: up to 5000 titles, client-side filter/sort (instant response)\nCareer stats: total works, average rating, career span, highest rated, most voted\nDecade breakdown with average ratings\n\nI load all data once (up to 5000 rows), then filter and sort client-side. No database round-trip for UI interactions.\n\nComparison Tools\nSide-by-side comparison (up to 4 people)\nCollaborators analysis (directors worked with, co-stars)\nSimilar movies (Jaccard similarity on genre/keyword vectors)\nDrill-down modals: click any title for full metadata, hyperlinks to IMDb\n\nThese patterns would work for many other customer and product analytics dashboards.\n\nDual Backend Setup\n\nThe app runs on two backends simultaneously - Hetzner (Germany) and Oracle Cloud (Singapore). Frontend has a toggle to switch between them.\n\nPrimary reason: Testing performance and as failover. If Hetzner goes down, flip the toggle to Oracle or can be automated at backend also. It's same FastAPI code & same database, just an additional VPS.\nFrontend sends backend=hetzner or backend=oci parameter\nServerless proxy routes to appropriate backend URL\nIn production, this toggle could move to backend config and trigger automatically on health check failures\n\nUseful for failover, A/B testing of infra changes and zero-downtime migrations.\n\nFeature Toggles\n\nI built several toggles into the architecture:\nAuth toggle (AUTH_ENABLED env var) - enable/disable Clerk authentication\nBackend toggle - switch between Hetzner/Oracle (exposed in UI for testing, would be backend-only in production)\nSmart Search toggle - user-facing checkbox to filter prolific people vs search all\nQuery timer toggle - show/hide execution times for each query\n\nThe query timer pairs with the backend toggle. Turn on the timer, switch backends, run the same query twice. You can see exactly how Hetzner compares to Oracle. I built it for debugging but kept it in the UI - to be able to do realistic comparisons.\n\nKey point - build these toggles into the architecture. You can decide later whether it's a user-facing feature or admin config. Easier to hide a toggle than to add one after launch.\n\nData Pipeline\n\nDashboard's fast because I pre-computed tables via the data pipeline. I create these tables directly on the server using the admin endpoint.\n\nPipeline scripts (in scripts-dataprocessing/ folder):\nDownload IMDb datasets (TSV files, ~10GB compressed)\nImport to DuckDB base tables (titlebasics, titleprincipals, namebasics, titleratings)\nBuild optimization tables:\nperson_filmography - denormalized (eliminates JOINs at query time)\nperson_stats - one row per person with pre-computed aggregates\nprolific_summary - leaderboards for all era/category combinations\ndashboard_cache - entire dashboard response as single JSON blob\n\nMost processing happens server-side via admin endpoint API calls. I send CREATE TABLE queries through the scripts to /api/admin/query/imdb.duckdb - SQL runs on the server. No need to download 10GB locally. I run the pipeline from my laptop, execution happens on the VPS.\n\nExample - creating person_filmography (91M rows):\n\n`python\nresponse = requests.post(\n f\"{BACKEND_URL}/api/admin/query/imdb.duckdb\",\n headers={\"Authorization\": f\"Bearer {ADMINAPIKEY}\"},\n json={\"sql\": \"\"\"\n CREATE TABLE person_filmography AS\n SELECT tp.nconst, tp.tconst, tp.category,\n tb.primaryTitle, tb.startYear, tb.titleType,\n tr.averageRating, tr.numVotes\n FROM title_principals tp\n JOIN title_basics tb ON tp.tconst = tb.tconst\n LEFT JOIN title_ratings tr ON tb.tconst = tr.tconst\n \"\"\"}\n)\n``\n\nRuns on the server, writes directly to the DuckDB file.\n\nRefresh cycle: I am yet to do a refresh. Planning to do the re-download and all table builds directly on the server instead of doing it via laptop.\n\nReusability\n\nThe patterns here aren't specific to movies and can be reused across dashboards. Frontend components extract cleanly. Backend endpoints are generic (POST SQL, get JSON). The optimization strategy - pre-compute what's expensive, filter what's big, cache what's static.\n\nBuilt for 230 million rows. Works the same for 230 thousand.\n\nContainer Memory Mismatch\n\nThis was a very new one for me. The initial Oracle Cloud deployment was 4x slower than Hetzner. Same code, same database, same container size (8GB). Earlier I thought it was because Oracle uses network-attached storage, Hetzner uses NVMe. The real problem was memory. Here's the issue:\n\nOracle's host machine has 24GB RAM. My container is limited to 8GB. When DuckDB starts, it checks how much RAM is available by reading /proc/meminfo. But containers don't have their own /proc/meminfo - they see the host's file. So DuckDB saw 24GB available, not 8GB.\n\nDuckDB's default is to use 80% of available RAM. 80% of 24GB is about 19GB. So DuckDB tried to allocate 19GB in an 8GB container.\n\nThe container didn't crash immediately. Instead, it started swapping - moving data between RAM and disk constantly. That's why queries were slow, not failing. Memory thrashing looks like disk latency from the outside.\n\nOn Hetzner, no problem. The host has 8GB RAM, container has no limit. DuckDB sees 8GB, uses about 6GB. Everything fits.\n\nFix: Set DUCKDBMEMORYLIMIT=7GB explicitly on Oracle. DuckDB ignores /proc/meminfo and uses your number instead. Leave 1GB for the OS and FastAPI.\n\nLesson: In containers, always specify DuckDB memory explicitly.\n\nResources\nLive Dashboard: https://www.tigzig.com/movie-explorer\nFrontend Code: github.com/amararun/shared-imdb-dashboards\nBackend Code: github.com/amararun/shared-duckdb-dashboards-backend\n\n2026 Infra Guide Series\nPart 1 - AI Coder\nPart 2 - Deployment & Hosting\nPart 3 - Security\n\n\nRelated\n\nTools: Movie Explorer, BRIQ, DATS-4 Database AI Suite\n\nExplore: Database AI apps, APIs and MCP, Build guides\n\nMore posts: From 12 second queries to under 1s: Optimizing a 230 Million Row Dashboard - 14 Bottlenecks I Had to Fix, CinePro - 230M Rows, 16GB Database, Instant Queries with DuckDB, How I Built a Sub-Second Movie Similarity Engine With a 10-Line SQL Query, Full backend Infrastructure for DuckDB production and analytics system, Your API Key Is Visible in the Browser. Even if you put it as Vercel's 'secret' backend env variable."
},
{
"slug": "database-ai-built-for-day-to-day-work-five-categories-ten-micro-apps-live-open-source-free",
"title": "Database AI, built for day-to-day work. Five categories, ten micro apps. Live, open source, free.",
"tags": [
"database-ai",
"text-to-sql"
],
"content": "Database AI, built for day-to-day work. Five categories, ten micro apps. Live, open source, free.\n\nWhat they do\nChat with Postgres, MySQL, and DuckDB in natural language\nConvert plain English to SQL and run it\nAnalyze data, create and transform tables\nVisualize with charts\nConnect on the fly or use pre-configured connections\n\n5 Categories of Database AI apps\n\nCustom builds, Rapid Deploy, ChatGPT, Realtime Voice & xlwings Lite\n\nWhat are Micro Apps\n\nTools built by end users, for themselves and small teams. Not company-level software.\n\nHow to build\nUse AI Coders like Cursor, Claude Code & Gemini CLI.\nFor React apps use platforms aistudio.google.com → Build to scaffold an app\n\nAre servers required?\nNot necessarily. Depends on the app\nSee the DuckDB browser database app. Full React app bundled into a standalone single HTML file. Available here: www.tigzig.com/sql-rooms\n\nArchitecture\n\nModular components to connect to any type of frontend and backend. Mix and match components to build what you need.\n\nLive App\n\nGo to www.tigzig.com → Database AI & SQL Apps\n\nResources\nHit Docs on app site for user guides, source codes and repos.\nVisit tigzig.com: detailed posts on using and building AI Micro Apps, single page React apps and using AI Coders.\nUse for free on app site or clone, customize and deploy at your end.\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals are in the PDF and slide images on the HTML page.\n\nwww.tigzig.com\n\nDatabase AI\n\n5 Categories\n\n10 Micro Apps\n\nLive | Open Source | Free\n\nChat with your PostGres, MySQL and DuckDB databases. Analyze. Transform. Visualize. All In natural language.\n\nBuilt for day-to-day work. For analysts and teams\n\nA set of AI tools that let you query, transform, and visualize databases directly in conversational English\n\nGenerally available features across all apps\n\nConvert natural language to SQL and execute query against your connected databases\n\nRun complex multi-step processing to create data marts and and transform tables with natural language\n\nConnect to databases on the fly, or set-up pre-configured connections.\n\nConnect to remote MySQL and Postgres databases, as well local in-browser with DuckDB\n\nWhat is Database AI\n\nTools made by analysts and end users\n\nIn finance, operations, marketing , analytics…\n\nFor day-to-day work\n\nFor use in small teams\n\nNot company level software\n\nWhat are Micro Apps\n\nApp Categories\n\nCustom BuildCustomized interface and functionalities\n\nRapid Deploy: With Flowise native user interface\n\nChatGPT: Connected to databases. Fast to deploy\n\nVoice AIRealtime voice AI connected to databases\n\nExcel to DatabaseNot an AI App. But a tool to connect to remote database from Excel. Runs Python in Excel with xlwings Lite\n\nCustom Build\n\nConnect to and chat with remote databases\n\nDATS-4: Remote Databases\n\nConnect to any database on-the-fly\n\nMulti-step multi-agent framework for reasoning-based advanced analysis\n\nChoice of 9 LLMs\n\nUpload CSV/TXT files directly to remote DB\n\nDownload files\n\nPython charts and statistical analysis\n\nDebugging logs\n\nCustom Build\n\nDuckDB running locally in your browser.\n\nDABX-1: DuckDB In-Browser\n\nCustom SQL Rooms implementation\n\nDuckDB runs locally in browser\n\nProcess Gigabytes files efficiently (CSV, TXT, Parquet, JSON)\n\nMultiple AI model support\n\nCharts and visualizations\n\nImport/export database files\n\nSQL editor for advanced queries\n\nChatGPT\n\nTransform ChatGPT into your database analyst.\n\nConnect ChatGPT to Any Database\n\nConnect to any MySQL or PostgreSQL\n\nFast deployment with no custom UI builds needed\n\nBuilt-in Python code interpreter for stats\n\nAuto-generated charts and visualizations\n\nCost-effective - no separate token costs\n\nFastAPI backend handles connections\n\nChatGPT\n\nChatGPT simultaneously connected to pre-configured database across multiple providers e.g Supabase, AWS, Neon, Aiven etc\n\nChatGPT : Fixed Databases\n\nDirect database connections to multiple pre-configured databases and schema\n\nBuilt-in ChatGPT Python code interpreter\n\nStatistical analysis and chart generation\n\nNo custom UI needed – ChatGPT is the UI\n\nFastAPI backend for connectivity\n\nChatGPT\n\nConnect ChatGPT to n8n for database queries, automation, and Google Apps Script integration.\n\nChatGPT - Connect via n8n\n\nUpdate any backend system via API endpoints\n\nAutomate tasks with Google Apps Script\n\nTransform and process data in n8n\n\nConnect to databases for real-time updates\n\nGenerate PDF reports and presentations\n\nRapid Deploy\n\nConnect to any database with Flowise UI.\n\nRapid Deploy: General Analyst\n\nRapid deploy Flowise user interface\n\nConnect to any database on-the-fly\n\nChat, analyze, transform data\n\nPython charts and statistical analysis\n\nGenerate formatted security analysis reports in PDF & HTML formats\n\nRapid Deploy\n\nAdvanced Database Analyst with Flowise UI\n\nRapid Deploy: Advanced Analyst\n\nRapid deploy Flowise user interface\n\nMulti-agent, multi-step reasoning based processing for advanced analysis.\n\nConnect to any database on-the-fly\n\nPython charts and statistical analysis\n\nRealtime Voice\n\nRealtime voice connected to databases. Rapid deploy with ElevenLabs widget\n\nRealtime Voice AI: Widget\n\nReal-time voice interaction with databases\n\nUses ElevenLabs rapid deploy widget\n\nConnected to pre-configured databases\n\nNote: This is an experimental app developed in 2024. Capacity permitting, a refurbished and updated version will be shared soon\n\nRealtime Voice\n\nRealtime voice connected to databases via WebRTC for granular customization of UI / UX\n\nRealtime Voice AI: WebRTC\n\nReal-time voice interaction with databases\n\nConnected to pre-configured databases\n\nConnected via OpenAI realtime API\n\nNote: This is an experimental app developed in 2024. Capacity permitting, a refurbished and updated version will be shared soon\n\nxlwings Lite\n\nNot an AI App, but an automation tool to connect to remote databases from within Excel\n\nExcel to DB: with xlwings Lite\n\nConnect to any PostGres and MySQL database from within Excel\n\nRuns Python script within Excel with xlwings Lite\n\nConnection via backend FastAPI server\n\nCore Architecture\n\nModular architecture connecting across different user interfaces and apps\n\nComponents Based Architecture\n\nGitHub Repos | Description\n\nMain App | The main application with the UI\n\nFastAPI: Database Connector | Handles Text-to-SQL processing, including file uploads\n\nFastAPI: Neon DB Creation | Temporary database creation with Neon\n\nFlowise Agent Schemas | Sequential Agent Framework with LLM Agent built with Flowise AI\n\nProxy Server | For API Calls to OpenAI / Gemini / Openrouter\n\nMCP Server - Markdown to PDF | For converting markdown to PDF\n\nQuant Agent Backend Repos | The TIGZIG Quants Agent app integrated into a single tab\n\nMix and match deployment of individual tools\n\nConnect components to your own user interface\n\nNumerous more components available open source at www.tigzig.com - web scraper, pre-formatted slide deck creator, Excel table to PDF, Excel and Google Sheets updater, file converters, finance data extractors…\n\nField Guides\n\nUser Guides & Source Code\n\nHit Docs on the app page\n\nLinks to source codes and build guide, including video guides\n\nAll source codes and user guide links on app page in docs section\n\nTIGZIG Analyst\n\nMicro-Apps for Analytics\n\nOpen Source\n\nwww.tigzig.com\n\nAmar Harolikar\n\nSpecialist - Decision Sciences & Applied AI\n\nBuilder of www.tigzig.com\n\nLive\n\nFree\n\n30+ Micro Apps for AnalyticsDatabase AI 🔹Quants 🔹 xlwings Lite 🔹 GPTs🔹MCP\n\nDatabase & SQL AI Apps\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, BRIQ, ChatGPT to Any Database\n\nExplore: Database AI Apps hub, Database Query Tool (MCP), xlwings Lite Practice Lab\n\nMore posts: AI for Databases: Field Guide, Live Apps & Lessons, Connect Excel to a Database with Tens of Millions of Rows? Tigzig.com Has 15+ Apps and Guides for That Alone. Just Ask Your AI., BRIQ App: DuckDB AI in Browser - 500MB Files, 4M+ Records, No Database Setup, ChatGPT connected to your databases. One-click deployment instructions for AI Coders"
},
{
"slug": "database-ai-sql-agent-connect-to-any-database-on-the-fly-live-open-source",
"title": "Database AI & SQL Agent - Connect to any database on-the-fly. Live. Open Source",
"tags": [
"database-ai",
"text-to-sql"
],
"content": "Database AI & SQL Agent - Connect to any database on-the-fly. Live. Open Source\n\nA short demo of the core text-to-SQL workflow in my DATS-4 suite (open source). Connect to any database (MySQL/Postgres) on-the-fly, explore, and analyze with natural language.\n\nDATS-4 is live at www.tigzig.com\n\nPath: Database AI & SQL Apps -> DATS-4 (Database AI Suite -v4.)\n\nThe process is direct:\nDump your credentials into the text box. The format doesn't matter; a credential-parsing AI step standardizes it to a valid JSON object.\nStart asking questions in natural language.\nTransform the data: summarize, merge, create new tables, and export the results.\n\nThe agent flow is designed for full transparency. It shows its work: the complete reasoning, the SQL it generated, and the final results in tables and charts. It's an auditable process.\n\nLive app, source codes and field guide at link above\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, BRIQ, Database Query Tool\n\nExplore: Database AI Apps hub, MCP servers and agents, Browse all apps\n\nMore posts: Database & SQL AI: On-the-fly database transformation with natural language. Connect, transform, and export instantly., Database AI, built for day-to-day work. Five categories, ten micro apps. Live, open source, free., AI for Databases: Field Guide, Live Apps & Lessons"
},
{
"slug": "database-ai-sql-now-choose-you-llm-gpt-5-deepseek-qwen-3-thinking-live-open-source",
"title": "Database AI & SQL - Now choose you LLM: GPT-5, Deepseek, Qwen 3 Thinking. Live. Open Source.",
"tags": [
"database-ai",
"text-to-sql"
],
"content": "Database AI & SQL - Now choose you LLM: GPT-5, Deepseek, Qwen 3 Thinking. Live. Open Source.\n\nLive at: www.tigzig.com\n\nYou need top-tier reasoning? Use Claude Sonnet 4. Need the latest model for testing? Swap in GPT-5. Need a cost-effective workhorse? Run DeepSeek.\n\nWatch what happens at the 1:45 mark. The agent's SQL query fails. It gets a 500 error. And then it pulls sample records to 'see' the data, fixes the SQL and re-runs the job successfully.\n\nIt's a full-stack tool: on-the fly database connect, instant Postgres DB creation with automated schema detection, full agent reasoning traces, PDF outputs, and more.\n\nThis isn't a theory. It's built on 15+ months of live client ops across 9 deployments.\n\nDeployed variants: Full Stack, Rapid Deploy & Custom GPT\n\nField Guide: My practical learnings: security and agent setup to LLM costing and deployments.\n\nTest it live: upload your file, connect to your database or use the sample file and prompts on the site\n\nLive apps, field guide and source code available at: www.tigzig.com\n\nPath: Database AI & SQL Apps\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, BRIQ, ChatGPT to Any Database\n\nExplore: Database AI Apps, Database Query Tool (MCP), Browse All Apps\n\nMore posts: Qwen3 Max now live on DATS-4 SQL Agent Suite for Advanced Analysis Better than DeepSeek R1, closer to Claude Sonnet 4 - at a lower cost., Sonnet 4.5. Released yesterday. Now live on DATS-4 SQL Agent Suite. Solid upgrade, but more 4.2 than 4.5., AI for Databases: Field Guide, Live Apps & Lessons"
},
{
"slug": "database-sql-ai-on-the-fly-database-transformation-with-natural-language-connect-transform-and",
"title": "Database & SQL AI: On-the-fly database transformation with natural language. Connect, transform, and export instantly.",
"tags": [
"database-ai",
"text-to-sql"
],
"content": "Database & SQL AI: On-the-fly database transformation with natural language. Connect, transform, and export instantly.\n\nWith DATS-4 Database AI Suite (Open Source):\nConnect: to any database (MySQL/Postgres) on-the-fly\nInstruct: the agent to add derived variables or perform data transformations\nExecute: the agent writes the SQL and creates a new, transformed table and logs the entire process\nValidate: review agent reasoning, SQL and debugging carried out by the agent\nExport: the finished table to CSV for further analysis from right within the app\n\nDATS-4 also supports direct CSV uploads, fixed DB connections and temporary Postgres instance on-the-fly.\n\nThe live app, source code, and my full Field Guide are available at the links below.\n\nDirect to DATS-4: www.tigzig.com/analyzer\n\nMain Site (25+ Tools): www.tigzig.com\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, BRIQ, Database Query Tool\n\nExplore: Database AI Apps hub, API and MCP hub, Browse all apps\n\nMore posts: AI for Databases: Field Guide, Live Apps & Lessons, Database AI & SQL Agent - Connect to any database on-the-fly. Live. Open Source, Database AI, built for day-to-day work. Five categories, ten micro apps. Live, open source, free."
},
{
"slug": "db-mcp-both-formats-both-engines-aug2026",
"title": "T20 and ODI Ball by Ball Are Now on Both Engines, Postgres and DuckDB. Run Your SQL Directly on My Server.",
"tags": [
"database-ai",
"mcp-servers"
],
"content": "T20 and ODI Ball by Ball Are Now on Both Engines, Postgres and DuckDB. Run Your SQL Directly on My Server.\n\nBoth T20 & ODI cricket ball by ball now there on both engines - Postgres & DuckDB. Run your SQL directly on my server. Free, no auth, no key. Calculate strike rates by batter against pace and against spin, over by over run rates through the death overs, head to head records between two teams across a decade, dismissal patterns by bowler type, home and away splits by venue... etc .. etc\n\nAuto-refreshes twice a day.\n\nTip: most calls come in over GET, and the commonest error is still the URL rather than the SQL, so if a query comes back looking mangled, send it as a POST with a JSON body instead.\n\nDocs\nfor humans: db-mcp.tigzig.com/v1/redoc\nfor your AI: db-mcp.tigzig.com/v1/openapi.json\n\nCredit: data sourced from Cricsheet, ODC-BY 1.0.\n\nNo MCP required, just ask your AI worker\n\nIf you have a ChatGPT or Claude subscription - both come with a Cowork app - you can simply share the message / link above and ask it to explore the data for you. Also with a Gemini subscription which comes with Antigravity .. which is actually a full fledged AI Coder. Comes with the $20 / Rs. 1,950 subscription also.\n\nNo MCP required.. just ask like in a regular chat bot ... they will go and fetch the data directly via API .. copy it to your local drive, create excel, deck, reports for you also if you'd like.\n\nJust download their desktop app and login .. that's it ..\n\nFor more on using AI Workers, Tigzig AI Worker hub at tigzig.com/claude-for-analytics\n\nWhat is allowed, and what gets refused\n\nIt is read only. Writes, DDL, and anything that reads a file or a URL are refused. There are caps on functions that can generate rows from nothing or spin CPU without limit, and the refusal names the specific function so you can drop that one and retry rather than rewriting the query.\n\nSchema discovery is deliberately open. information_schema works, SHOW TABLES and DESCRIBE work, so you or your agent can find the columns without me publishing them anywhere.\n\nA couple of the guards were tighter than they needed to be in the first week and got loosened once I realized I was refusing many regular queries.\n\nThe Postgres now runs on my self hosted box, so way faster than earlier which was on a managed instance.\n\ndb-mcp.tigzig.com/redoc\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, BRIQ, QRep - Security Reports\n\nExplore: API and MCP catalog, Claude for Analytics hub, Database AI apps\n\nMore posts: Getting Errors Running SQL Against the Tigzig Cricket Database? The Most Common Failure Is the URL, Not Your SQL., Run Your Own SQL on My Cricket Database. T20 and ODI Ball by Ball Since 2002, Free, No Signup and No Key., A Database Server With API and MCP You Can Point an AI Agent At, With Ball-by-Ball Cricket Behind It., Four Days of an Open SQL Endpoint. What Real Cricket Queries Needed, and the SSRF Someone Found in the First Couple of Hours."
},
{
"slug": "db-mcp-cricket-ball-by-ball-api-aug2026",
"title": "A Database Server With API and MCP You Can Point an AI Agent At, With Ball-by-Ball Cricket Behind It.",
"tags": [
"database-ai",
"mcp-servers"
],
"content": "A Database Server With API and MCP You Can Point an AI Agent At, With Ball-by-Ball Cricket Behind It.\n\nLive now on TigZig, a database server with API / MCP you can point an AI agent at, with cricket ball-by-ball + match-master data behind it. Updated daily. From 2002 to Aug'26. About 1.36 million ODI deliveries and 786K T20 deliveries across roughly 6,000 matches, and the match table goes wider still at about 6,900 matches because it carries Tests as well.\n\nThe API / MCP exposes read-only SQL over two engines, Postgres for ODI and DuckDB for T20.\n\nIt is free and there is no sign-up, auth or key.\n\nThe pipeline is new and in live testing, so treat the data as provisional for now .. if you find any issue, drop a note on the Feedback button on tigzig.com. The data is sourced from cricsheet.org.\n\nAdd it to Claude, ChatGPT or Cursor as a connector, or have them call the REST endpoint directly.\n\nDocs\nHumans: tigzig.com/apis/database\nAI: db-mcp.tigzig.com/v1/openapi.json\nMCP endpoint, if you are adding it as a connector: db-mcp.tigzig.com/mcp\n\nA few things under the hood\nTwo engines, same schema. ODI sits on Postgres and T20 sits on DuckDB, same column names. DuckDB is running on my own server and it is fairly fast. Postgres is on a Supabase free tier for now, which is the experimental half of this, and I am watching how it gets used and what the latency and the error rows look like. Depending on that I will optimise it where it sits or move it to the self-hosted Postgres I already run.\nPutting SQL on the open internet is easy to build and as easy to get wrong. Read-only at the database level, timeouts, row caps, a ceiling on query complexity, protection against SQL injection and CPU exhaustion attacks, blocked system catalogs ...and then testing all of it.\nEverything here is public, so the limits are per IP, because that is the only handle there is, and I would ask people to stay polite to it.\nThe errors are like additional documentation...get a column wrong and the response hands back the real tables, the columns, the join key and a working call.\n\ntigzig.com/apis/database\n\nWhat the ball-by-ball data actually gives you\nThe T20 data starts on the first men's T20 international ever played, 17 Feb 2005, Australia v New Zealand. It's the complete history of the format, every delivery. The ODI side goes back to 2002.\nBall-by-ball is a very granular level...one-row-per-delivery - over-by-over chase pressure, who bowled to which batter, dot-ball sequences, phase-wise economy.\nThe ball table has no winner and no competition - the match table does. One join on match_id, and you have link to final scorecard and tournament and player details too.\n\nCricket data from Cricsheet, published under the Open Data Commons Attribution License 1.0 (ODC-BY).\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, BRIQ, QRep - Security Reports\n\nExplore: API and MCP catalog, Database AI apps, Claude for Analytics hub\n\nMore posts: VIGIL Data API and Download - 315K+ Records, 7 Tables, Updated Daily, Range Requests on the MFPRO NAV File Are Fixed. DuckDB and Polars Queries No Longer Get Refused Midway., Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users, Tigzig is Built AI-Agent First. 300+ Assets, and the Fastest Way In Is to Ask Your AI."
},
{
"slug": "db-mcp-cricket-sql-queries-aug2026",
"title": "Run Your Own SQL on My Cricket Database. T20 and ODI Ball by Ball Since 2002, Free, No Signup and No Key.",
"tags": [
"database-ai",
"mcp-servers"
],
"content": "Run Your Own SQL on My Cricket Database. T20 and ODI Ball by Ball Since 2002, Free, No Signup and No Key.\n\nCricket data analysts, now run your own SQL on my cricket database, free, with no signup and no key. T20 & ODI Ball by ball since 2002, 2M+ deliveries, 6000+ matches, PostGres & DuckDB over API or connect your agent via MCP.\n\nMatch level detail like winner, venue, toss and player of the match in a second table you join on match_id, so whether a ground favours batting first, or how a batter goes against a particular bowler, is a single query.\n\nThe service has been live for a week and usage is growing faster than I was expecting... it went up as an experiment on a small quota and the latencies show it. Some queries have been timing out under load and I have been optimizing performance. Next - both engines are moving to my dedicated API server over the next few days, with the query performance and concurrency tuned properly. Same endpoints throughout, nothing changes at your end.\nAPI Docs (agents): db-mcp.tigzig.com/v1/openapi.json\nAPI Docs (humans): db-mcp.tigzig.com/redoc\n\nMore questions the same two tables answer\n\nOther than the ones on the image, a few more that come out of the same two tables.\nHow a batter actually gets out, the caught, bowled and lbw split, and whether it shifts against a particular bowler\nWhich sides give away the most in wides and no balls, usually a bigger number than people expect\nWhether winning the toss is worth anything at a given ground, since the toss call and the result are both in there\nHow scoring rates have moved season by season since 2002\nWorld Cup and tournament matches against plain bilateral series, using the event column\nAssociate nations against the Full Members, which is a column in the match table rather than something you have to guess at\nAnd the officials are in there too, both umpires, the TV umpire and the match referee, if that is the sort of thing you want to dig into\n\nA few things on the data side\n\nmatchinfo carries every format, so it holds more matches than either ball table does. Roughly 6,900 rows in there, of which about 2,600 are ODI and about 3,500 are T20, and the remaining 900 or so are Tests where the match level detail is present but the ball by ball is not. Join on matchid and filter on match_type, otherwise the denominator is wrong and the answer still looks reasonable.\n\nThe ball tables are format scoped per engine, ODI on Postgres and T20 on DuckDB, so no single query spans both, you run it twice and compare.\n\nIt is men's internationals, and that includes the Associate nations, so a plain top run scorers over the whole ball table pools every level of international cricket together. team1icctype and event sit in match_info if you want to narrow it.\n\nResults cap at 1000 rows and the response carries a truncated flag, so check that before treating a leaderboard as the full list.\n\nAnd the schema is small on purpose, two tables per engine, so information_schema or DESCRIBE will tell you everything there is rather than you guessing at column names.\n\nCricket data from Cricsheet, published under the Open Data Commons Attribution License 1.0 (ODC-BY).\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, BRIQ, QRep - Security Reports\n\nExplore: API and MCP catalog, Database AI apps, Claude for Analytics hub\n\nMore posts: A Database Server With API and MCP You Can Point an AI Agent At, With Ball-by-Ball Cricket Behind It., Four Days of an Open SQL Endpoint. What Real Cricket Queries Needed, and the SSRF Someone Found in the First Couple of Hours., Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users, VIGIL Data API and Download - 315K+ Records, 7 Tables, Updated Daily"
},
{
"slug": "db-mcp-four-days-ssrf-query-guards-aug2026",
"title": "Four Days of an Open SQL Endpoint. What Real Cricket Queries Needed, and the SSRF Someone Found in the First Couple of Hours.",
"tags": [
"database-ai",
"security"
],
"content": "Four Days of an Open SQL Endpoint. What Real Cricket Queries Needed, and the SSRF Someone Found in the First Couple of Hours.\n\nI put the Cricket database API / MCP live about four days ago, ball by ball data, open endpoint to run your SQL queries on my database. No key and no login. 650+ successful API calls since then and the two types of traffic - real cricket queries and exploitation attempts.\n\nI have changed a lot of things ... all of it came out of reading what callers were actually being refused. Two guard rules were turning away about 62% of real queries. The join limit was set so low that ordinary two table work broke, schema discovery was blocked, and the errors now tell you what to fix instead of just saying no. Plus many more tweaks to remove bottlenecks and speed up processing.\n\nAt the same time, within the first couple of hours someone found an SSRF, server side request forgery, which is like getting my server to make network calls on their behalf. It worked. My read-only database connections had external access left switched on, which was a hole at my end. It came up in monitoring about ten minutes later and was closed inside the hour. I have since hardened the engine itself.\n\nThe docs have been totally refurbished and updated with guidelines on what is allowed and what is blocked so it's easier at your end to setup and debug queries.\nFor humans: db-mcp.tigzig.com/redoc\nFor agents: db-mcp.tigzig.com/v1/openapi.json\n\nHow cricket analysts are using it\n\nA general sense of how cricket data analysts are using the API: scorecards rebuilt from the raw deliveries, death over analysis, bowler dismissal analysis, economy and strike rates, whether the toss changes anything, partnerships, batting first against chasing, phase of innings splits. And then head to head records, and the cuts by opposition, by year and by season.\n\nSome quick things from the docs\nUp to 10 joins per select, subqueries three deep, and window functions are allowed, so partnerships, running totals and rank within an innings all work.\nTwo traps. team1icctype and event sit in matchinfo, not in the ball table, so a plain top run scorers query pools every level of international cricket and Associate nations come up alongside Full Members. And matchinfo carries all three formats in both engines while each ball table holds only one, so match counts will not reconcile unless you filter on match_type.\nScope is men's internationals only for now, no domestic or franchise cricket.\nAnd two small ones. Every query caps at 1000 rows with a truncated flag, so set your own limit and a false on truncated is your proof nothing was cut. TSV output is way smaller than JSON if you are handing results to an agent.\n\nA bit more on the SSRF\n\nIt was not a single request. It ran for around an hour and came from around 159 addresses, almost all used once, across 32 networks, two thirds cloud and the rest home broadband and other providers. Dozens of attempts were refused before anything went through, including every file and URL reader function, an attempt to disguise a function name, and eleven separate attempts to switch a security setting off. What eventually worked was a string evaluation path with the blocked keyword assembled at runtime, so the filter never saw the word it was looking for. That class of bypass is closed now, at more than one level. What it reached was a public file. Nothing was aimed at my data, and nothing internal was touched.\n\nCricket data from Cricsheet, published under the Open Data Commons Attribution License 1.0 (ODC-BY).\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, BRIQ, QRep - Security Reports\n\nExplore: API and MCP catalog, Security checklist, Database AI apps\n\nMore posts: A Database Server With API and MCP You Can Point an AI Agent At, With Ball-by-Ball Cricket Behind It., Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users, Zero Callers Rate Limited in 24 Hours. 6,500+ Calls and 9.7 GB Served, After the TigZig API Limits Went Up., VIGIL Data API and Download - 315K+ Records, 7 Tables, Updated Daily"
},
{
"slug": "db-mcp-test-cricket-aug2026",
"title": "Test Cricket Is Now Live in the Tigzig Open Cricket Database, Alongside T20 and ODI. 3.8 Million Deliveries.",
"tags": [
"database-ai",
"mcp-servers"
],
"content": "Test Cricket Is Now Live in the Tigzig Open Cricket Database, Alongside T20 and ODI. 3.8 Million Deliveries.\n\nTest cricket is now live in the Tigzig open cricket database, along with T20 and ODI, ball by ball, DuckDB and Postgres, read only SQL access via API and MCP. ~6,900 matches, 25 yrs+, 3.8 million deliveries. Sourced from Cricsheet.\n\nREST API + MCP server.\n\nData is refreshed twice a day.\n\nTest goes back to 2001, ODI to 2002, T20 to 2005.\n\nNo key, no auth and free.\n\nDocs\nhumans: tigzig.com/apis/database\nAI: db-mcp.tigzig.com/v1/openapi.json\n\nCredit: data source Cricsheet, ODC-BY 1.0.\n\nWhat is in there\n\nThree ball-by-ball tables, one per format, plus a matchinfo table that carries venue, dates, teams, toss and result. They all join on matchid. Every delivery has the striker, the non-striker, the bowler, the runs off the bat, the extras broken out, and the wicket if there was one.\n\nBecause it is ball by ball and not pre-aggregated, anything you can write in SQL over deliveries is open to you. Strike rate against pace versus spin, run rates through the death overs, how a particular ground behaves batting first, head to head between two teams over a decade.\n\ntigzig.com/apis/database\n\nOne thing I added with Test\n\nThere are now overno and deliveryinover columns, both plain whole numbers. The older ball column packs the over and the delivery into one decimal, and it has a catch: the tenth delivery of an over is written 1.10 at the source, which as a decimal is the same value as 1.1, the first delivery. So in long overs two rows share a value and sorting by it puts them in the wrong order. The new columns cannot run into that. Order by innings, overno, deliveryinover and you get the innings as it was bowled.\n\nAn over goes past six deliveries more often than you would think, because a wide or a no-ball has to be bowled again. Eight and ten ball overs are ordinary, and the longest in this data ran to 15.\n\ntigzig.com\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, BRIQ, QRep - Security Reports\n\nExplore: Database API docs, API and MCP catalog, Database AI apps\n\nMore posts: T20 and ODI Ball by Ball Are Now on Both Engines, Postgres and DuckDB. Run Your SQL Directly on My Server., Have Your AI Analyze the Cricket Data on Tigzig. 24+ Years of Ball by Ball, ODI and T20, Across 6000+ Matches., Run Your Own SQL on My Cricket Database. T20 and ODI Ball by Ball Since 2002, Free, No Signup and No Key., Getting Errors Running SQL Against the Tigzig Cricket Database? The Most Common Failure Is the URL, Not Your SQL."
},
{
"slug": "db-mcp-url-encoding-errors-aug2026",
"title": "Getting Errors Running SQL Against the Tigzig Cricket Database? The Most Common Failure Is the URL, Not Your SQL.",
"tags": [
"database-ai",
"mcp-servers"
],
"content": "Getting Errors Running SQL Against the Tigzig Cricket Database? The Most Common Failure Is the URL, Not Your SQL.\n\nAre you running SQL against the Tigzig cricket database? Getting errors? The most common failure has nothing to do with your SQL, it is the URL it travelled in.\n\nYou can send a query to us two ways. Inside the URL, which is a GET, or inside a JSON body, which is a POST. If you use the URL, the whole query has to be percent encoded first, and that is the step where it goes wrong. A plus sign inside a URL already means a space, so SUM(runsoffbat + extras) reaches us with the plus gone. Line breaks can get eaten, so two words end up stuck together. Some clients stop encoding partway and the query arrives cut off. The SQL on your screen is correct, and by the time it gets here it is a different query.\n\nFor these URL problems, the error response you get names the cause where it can .... and for the rest it quotes your SQL back exactly as we received it so you can see what changed.\n\nWhy am I keeping GET? It is the only thing a plain browser can do, agent driven browsers and no code HTTP nodes use it, and those calls work perfectly well too.\n\nI added those to the docs ...the most common cases and what to send instead. Doc links below, comments have some quick points.\nFor humans: db-mcp.tigzig.com/v1/redoc\nFor your AI: db-mcp.tigzig.com/v1/openapi.json\n\nData from: Cricsheet, published under the Open Data Commons Attribution License 1.0 (ODC-BY).\n\nA few quick points\nJust to note that ODI sits on Postgres, T20 on DuckDB. The two engines are different SQL dialects, so a query that works on one can fail on the other.\nThere is a 1000 row cap.. for more than that you get the first 1000 and truncated: true. Always read that field.\nSchema discovery, SHOW TABLES and DESCRIBE both work.\nIf you set no limit at all, one is added for you. Everything else behaves normally: ORDER BY, OFFSET, CTEs and aggregates are unaffected, so LIMIT ... OFFSET ... pagination works.\nSet your own limit. It is faster, and a false on truncated is your proof nothing was cut.\n\nWhat it looks like when it goes wrong\nSUM(runsoffbat + extras) arriving as SUM(runsoffbat extras). The plus became a space. Send it as %2B.\nAS runs FROM t20ballbyball arriving as AS runsFROM t20ballbyball. A line break got eaten.\nSELECT gender, COUNT(*) FROM ... arriving as SELECT gender, COUNT. The encoder gave up at the bracket.\n\nIf you would rather not think about any of this, send a POST with a JSON body and there is no encoding step to get wrong. Check updated docs below.\nFor humans: db-mcp.tigzig.com/v1/redoc\nFor your AI: db-mcp.tigzig.com/v1/openapi.json\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, BRIQ, QRep - Security Reports\n\nExplore: API and MCP catalog, Database AI apps, Claude for Analytics hub\n\nMore posts: Run Your Own SQL on My Cricket Database. T20 and ODI Ball by Ball Since 2002, Free, No Signup and No Key., A Database Server With API and MCP You Can Point an AI Agent At, With Ball-by-Ball Cricket Behind It., Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users, Four Days of an Open SQL Endpoint. What Real Cricket Queries Needed, and the SSRF Someone Found in the First Couple of Hours."
},
{
"slug": "diesel-above-six-dollars-sep2026",
"title": "US Diesel Is Above $6 a Gallon for the First Time, and the Cost Is Already Moving Through the Economy",
"tags": [
"portfolio-quants"
],
"content": "US Diesel Is Above $6 a Gallon for the First Time, and the Cost Is Already Moving Through the Economy\n\nUS diesel is above $6 a gallon for the first time. It was $6.53 on 21 September, up 74% on a year ago, and the cost is already moving through the economy.\n\nDiesel powers the trucks, trains and farm machines that carry almost everything, so when it goes up the cost reaches household bills and company margins. Truck freight rates are up 14% on a year. Airfares are up 23%, and jet fuel comes from the same part of the barrel as diesel. Heating oil is forecast to cost up to 31% more this winter if prices hold. Farmers are paying for it at harvest, and small truckers most of all, because they have the least room to pass it on.\n\nThe squeeze is in refining. Crude is off its April peak, but diesel now costs about $3.60 a gallon more than Brent crude, wider than at any point in 2022. The IEA expects Russian refining to stay about 30% below its pre-war level for the next 18 months, after drone strikes and sanctions, and the Iran war has hit refineries in the Gulf.\n\nOn how long this could last: the IEA says the drone strikes now hit the secondary units of Russian refineries, which take six to eight months to replace, and sanctions make repairs harder. Its September report also shows refining margins in the Atlantic Basin at record levels in August, led by diesel. So a ceasefire would not bring diesel down quickly.\n\nWashington is now weighing a ban on diesel exports. Trump has backed one and says a decision will come soon, the Treasury is checking whether it would work, and the oil industry says it would raise prices. As of this week no decision has been made.\n\nThis comes on top of pressure that was already building. The US 10-year yield closed above 5% this month for the first time since 2007, the job market looks calm only on the headline numbers, and the Fed raised rates on 16 September with inflation still above its target.\n\nEarlier pieces in this series\nBond markets have started acting\nUS jobs, calm before the storm\nThe Fed is caught\n\nAll my earlier analyses: tigzig.com/analysis\n\nSources\n\nSources for the one-pager are EIA, BLS via FRED, the IEA's September Oil Market Report and TREMOR.\n\nIEA on Russian refining: Russian refining sector struggles amid intensifying Ukrainian attacks\n\nIEA Oil Market Report, September 2026\n\nCNBC on how diesel costs spread through the economy: Diesel, gas prices and the US economy\n\nThe diesel and inflation series behind the one-pager are on tigzig.com/tremor.\n\nUS diesel is above $6 a gallon for the first time, up 74% on a year ago, and the cost is already moving through the economy"
},
{
"slug": "duckdb-isn-t-just-fast-sql-it-s-python-sql-and-compression-all-in-one-box",
"title": "DuckDB isn't just fast SQL. It's Python, SQL and compression all in one box.",
"tags": [
"duckdb"
],
"content": "DuckDB isn't just fast SQL. It's Python, SQL and compression all in one box.\n\nDuckDB\n\nI thought DuckDB was fast SQL with great compression. Speeds are lightning fast - I was happy with that. Then I went through Jasja De Vries's 30-day DuckDB series. Day by day. All 30.\n\nTurns out I was using maybe 10% of what it can do.\n\nMy analyst and data scientist friends, DuckDB isn't just fast SQL. It's Python, SQL and compression all in one box.\n\nFeatures I didn't know existed\n\n SELECT EXCLUDE - SELECT EXCLUDE (col1, col2) - grab everything except specific columns. SQL can't do this\n Prefix-aliasing - total: price qty instead of price * qty AS total - name comes first. Reads left-to-right.\nReusable aliases - Define alias in SELECT, use it in WHERE, GROUP BY, ORDER BY. SQL forces you to repeat the expression.\nLIMIT with % - LIMIT 10% instead of row counts. SQL doesn't have this.\nQUALIFY - Filter on window functions directly. SQL requires a nested subquery.\nBY ALL - GROUP BY ALL, ORDER BY ALL - SQL requires listing every column.\n LIST comprehensions - [x2 FOR x IN scores IF x > 70] - Python syntax inside SQL.\nLambda functions - list_filter(arr, x -> x > 10) - SQL has no lambdas.\nDot operator chaining - price.CAST(FLOAT).ROUND(2) - method chaining like Python.\n Glob patterns - SELECT FROM 'logs/*.csv' - query 1000 files with one line.\n Direct file queries - SELECT FROM 'data.parquet' - no CREATE TABLE needed.\n\nIf you work with SQL, SAS, or Python for analytics - this series from Jasja fills gaps you didn't know you had.\n\nFull 30-day series: https://eli5.eu/blog/index.html\n\nAmazing work from Jasja.\n\n\nRelated\n\nTools: BRIQ, DUCKIT - CSV to DuckDB, DATS-4 Database AI Suite\n\nExplore: Database AI Apps hub, Database Query Tool (MCP), Full app catalog\n\nMore posts: How I Built a Sub-Second Movie Similarity Engine With a 10-Line SQL Query, Full backend Infrastructure for DuckDB production and analytics system, From 12 second queries to under 1s: Optimizing a 230 Million Row Dashboard - 14 Bottlenecks I Had to Fix"
},
{
"slug": "duckdb-meets-excel-xlwings-lite-data-tools",
"title": "DuckDB Meets Excel: xlwings Lite Data Tools",
"tags": [
"duckdb",
"xlwings-lite",
"python-in-excel"
],
"content": "---\n\nExcel: xlwings Lite Data Tools\" \nslug: duckdb-meets-excel-xlwings-lite-data-tools\ndate_published: 2026-01-09T06:27:00.000Z\noriginal_url: https://www.tigzig.com/post/duckdb-meets-excel-xlwings-lite-data-tools\nsource: fresh\nprocessed_at: 2026-01-09T06:27:00.000Z\n\nDuckDB Meets Excel: xlwings Lite Data Tools\n\nxlwings Lite runs Python in Excel - you get DuckDB, Pandas, ML libraries right inside Excel. The constraint: it can't access your local file system due to browser restrictions.\n\nThat's a problem for large datasets. You have a 1.2 GB DuckDB file on GitHub. Or a 1.5 GB CSV from a client. Or a SQLite database or a JSON dump....\n\nxlwings Lite can process them - but how do you get them in?\n\nI built two tools to solve this.\n\nTool 1: DuckIt - File Converter & Hosting\n\nDrop CSV files, get back DuckDB or Parquet with shareable link. Paste that link into xlwings Lite, start querying. Tested with 1.5 GB CSVs, 10M+ records. Or upload existing DuckDB/Parquet for instant sharing.\n\nDuckIt Diagram\n\nTool 2: xlwings Lite Data Importer\n\nPulls files from URLs directly into Excel. Paste URL, click import. Works with DuckIt, GitHub, Google Drive, Dropbox.\n\nHandles DuckDB, SQLite, CSV, Parquet, JSON. Auto-detects file type, loads directly into DuckDB. Parquet remains as is. Tested up to 1.2 GB DuckDB download, 20M+ records.\n\nxlwings Downloader\n\nExample Workflow\n\nClient sends 1 GB CSV. Upload to DuckIt get DuckDB link paste in Excel query with SQL. No email attachments, no local file management.\n\nDuckIt xlwings\n\nTechnical Details\n\nDuckIt does CSV-to-Parquet conversion in-browser, backend creates DuckDB with time-limited signed URLs. Importer runs in xlwings Lite with CORS bypass via Cloudflare Worker. Multi-level file type detection (URL header magic bytes).\n\nBoth tools open source. Full technical details, how-to, and live demo files at links below.\n\nPrivacy Notice\n\nFiles stored on my server with time-limited signed URLs. API calls logged. Directory access is token authenticated. Free service, no guarantees. Deploy your own for production use.\n\nLinks\nDuckIt: tigzig.com/duckit-xlwings\nxlwings Lite Data Importer: Download Link\n\nBuilt with xlwings Lite by Felix Zumstein\n\nxlwings Lite official site: lite.xlwings.org\n\n\nRelated\n\nTools: DUCKIT - CSV to DuckDB, Data Importer, Practice Lab\n\nExplore: AI in Excel hub, Database AI apps, Browse all apps\n\nMore posts: xlwings Lite Data Importer v2 Released, xlwings Lite Local File Access: 8 Patterns You Can Use Today, Talk to Your Database from Excel - Postgres, DuckDB - via Claude in Excel with MCP, BRIQ App: DuckDB AI in Browser - 500MB Files, 4M+ Records, No Database Setup"
},
{
"slug": "ecb-npl-small-bank-corporate-strain-jun2026",
"title": "Euro-Area NPLs Look Calm. Split by Bank Size, the Small-Bank Corporate Book Is at 5.16% and Climbing.",
"tags": [
"portfolio-quants"
],
"content": "Euro-Area NPLs Look Calm. Split by Bank Size, the Small-Bank Corporate Book Is at 5.16% and Climbing.\n\nPublished: 2026-06-19\n\nEuro-area non-performing loans are near a decade low - but the calm headline hides where the stress is. Split by bank size and the small banks (less significant institutions, LSI - the Sparkassen, cooperative and savings banks; ~15% of euro-area bank assets) have pulled ahead of the big banks (significant institutions, SII, ~85% of assets). Big-bank NPLs sit near 2%; the small-bank corporate book has climbed to 5.16%, up about a third since 2023 and still rising. The consumer book is steady; the corporate book is the engine, and within that the two stressed corners are commercial real estate and small-business loans. Germany, Austria and Italy carry most of it - right where corporate insolvencies are running hot (Germany +57% vs the 2016-19 pre-pandemic May average per IWH; Austria 6,810 full-year 2025, +3.4%, property failures +35%; Italy ~+38% YoY per Allianz Trade). Households stay calm in the big systems (Germany 1.6% NPL, Netherlands 0.7%); the euro-area signal is corporate. Supervisors agree (ECB SREP 2025, OeNB FSR 51, ECB FSR May 2026) and tailored letters have gone to banks. Cycle to watch, not a system in trouble. 10-slide deck pulled live from the ECB SDMX REST API on 20 Jun 2026 (Q1 2026 release for SII / Q4 2025 for LSI - LSI lags one quarter by design, so SII vs LSI comparisons all use Q4 2025). Full 8-table copy-as-markdown dataset published at tigzig.com/ecb-npl-data-tables. Plus the parallel for the US: bank headline is also calm but the strain sits on the consumer (cards + autos) and in private credit, where losses hit the P&L directly and never show as a bank charge-off.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/ecb-npl-small-bank-corporate-strain-jun2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/ecb-npl-small-bank-corporate-strain-jun2026 with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/ECBNPLNOTE_JUN2026.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nSlide 1 - A calm number, hiding where the stress is\n\nEuro-area banks · non-performing loans · released 19 June 2026.\n\nEuro-area non-performing loans are near a decade low. But split the big banks from the small, and the small banks have pulled ahead.\n\nThe small banks have gone from cleaner than the big banks to worse, and the gap is widening.\n\nSource: ECB Supervisory Banking Statistics, via the ECB SDMX API.\n\nSlide 2 - The stress is on the corporate side\n\nSplit the small banks' book and only one half is moving.\n\nThe consumer book is steady. The corporate book is the engine.\n\nSource: ECB supervisory banking data, via the ECB SDMX API, Q4 2025.\n\nSlide 3 - What is under stress: property and small firms\n\nTwo corporate corners carry the stress: commercial real estate and small-business loans.\n\nCommercial real estate and small-business lending: the two corners under the most pressure.\n\nRates: ECB supervisory data (SDMX API), Q4 2025. Context: ECB SREP 2025; OeNB FSR 51 (Apr 2026).\n\nSlide 4 - In the supervisors' own words\n\nThe qualitative read from the supervisors matches the numbers.\n\nCommercial-property issues \"continued to weigh on asset quality, particularly in jurisdictions with historically low NPLs such as Germany, Austria, Estonia and Luxembourg.\" - ECB, Supervisory Review (SREP) 2025\n\n\"SME exposures remained a source of concern, as corporate insolvencies... remained elevated... particularly for highly leveraged firms and export-oriented sectors (automotive, chemical, pharmaceutical).\" - ECB, Supervisory Review (SREP) 2025\n\n\"The loan quality of small and medium-sized enterprises and commercial real estate companies remains strained.\" - OeNB (Austria's central bank), Financial Stability Report 51, April 2026\n\nSupervisors have acted: tailored letters went to banks, with most measures due by end-2026.\n\nSlide 5 - But keep it in proportion\n\nThe small banks are only 15% of euro-area bank assets, so the average hides them.\n\n15% of the system, but this is where the corporate stress is concentrated.\n\nSlide 6 - Where it lives\n\nA few specific systems, not simply the big countries.\n\nHouseholds stay calm in the big systems (Germany 1.6%, Netherlands 0.7%); only small markets like Greece run hot. The euro-area signal is corporate.\n\nSources: ECB supervisory banking data (SDMX API), Q4 2025; Banca d'Italia FSR No.2 2025.\n\nSlide 7 - The backdrop: more firms failing\n\nCorporate failures have turned up across exactly the three systems showing the bank stress.\n+57% Germany: May 2026 corporate insolvencies vs the 2016-19 pre-pandemic May average (IWH early indicator)\n6,810 Austria: full-year 2025, +3.4% on 2024 and a fresh high; property-sector failures +35%\n+38% Italy: 2025 versus 2024, among the steepest rises in Europe (Allianz Trade)\n\nFailures are rising across all three stress pockets, with property the common thread. A cycle to watch, rather than a system in trouble.\n\nSources: IWH Insolvenztrend (May 2026); Reuters/Destatis (Dec 2025); KSV1870 (Jan 2026); Allianz Trade (Oct 2025); ECB FSR (May 2026).\n\nSlide 8 - Data and coverage\n\nWhere the numbers come from, and the one timing rule that governs the comparisons.\n\nData source\n\nExtracted via the ECB SDMX REST API (data-api.ecb.europa.eu/service/data/SUP/) on 20 June 2026, and cross-checked against the ECB Supervisory Banking Statistics quarterly publications. The publication links on the Sources page are the human-readable equivalent of the same SDMX records.\n\nFrequency and timing\n\nThese supervisory series are quarterly, the highest frequency available. Big-bank (SII) data runs to Q1 2026, released 19 June 2026. Small-bank (LSI) data is published one quarter later by design, so the latest is Q4 2025 (LSI Q1 2026 is due 16 July 2026). Every SII-versus-LSI comparison therefore uses the same quarter, Q4 2025.\n\nCoverage\n\nSII and LSI together are the whole euro-area supervised banking system, about 1,920 banks and EUR 33.8 trillion of assets under the Single Supervisory Mechanism (109 significant banks at about 85% of assets, roughly 1,811 less-significant ones at about 15%). This is the complete sector, not a sample.\n\nA few bank-like activities sit outside it, branches of non-EU banks, standalone investment firms and money-market funds, but they are minor for a loan-book and NPL view.\n\nSegments, country view, and OeNB\n\nCRE and SME NPL ratios are published at the big-bank level; the small-bank split is not published euro-area-wide, so Austria (OeNB, the Oesterreichische Nationalbank) is used as a concrete small-bank example. Country book size uses published LSI total assets; euro NPL volumes are not estimated.\n\nNot just an API call, the full dataset is published. Every figure here is downloaded, cleaned and organised into eight tables (seven from the ECB API, plus the IWH insolvency series), each copyable as markdown in one click: tigzig.com/ecb-npl-data-tables.\n\nSlide 9 - Definitions\n\nWhat the numbers mean, in the ECB's own terms.\n\nNon-performing loan (NPL)\n\nA loan is non-performing when the borrower is \"unlikely to repay\", or when \"more than 90 days have passed\" without payment. It is a stock measure, the outstanding balance of non-performing loans as a share of gross loans (excluding central-bank cash), not a charge-off or loss rate. Comparable to a gross non-performing-asset ratio. (ECB and EBA definition.)\n\nBig banks (significant institutions, SII)\n\nThe largest banking groups, supervised directly by the ECB, about 85% of euro-area bank assets. In the composition we combine the published categories: global systemically important banks (G-SIBs), universal and diversified lenders, retail and consumer lenders, and other significant institutions.\n\nSmall banks (less significant institutions, LSI)\n\nSmaller banks supervised by national authorities under ECB oversight, about 15% of euro-area bank assets. In several countries these are savings and cooperative banks.\n\nSlide 10 - Sources\n\nECB and national primary data; every figure cross-checked to the published source.\n\nECB Supervisory Banking Statistics\n\nThe bank-size and sector NPL data, the spine of this note. Pulled via the ECB SDMX API; Q1 2026 release for the big banks (19 June 2026), Q4 2025 for the small banks. bankingsupervision.europa.eu · data.ecb.europa.eu\n\nECB SREP 2025 · OeNB Financial Stability Report 51 · ECB FSR May 2026\n\nCRE stress \"particularly in Germany, Austria, Estonia, Luxembourg\"; SME \"a source of concern\" (ECB SREP, Nov 2025). Austria small-bank CRE 8.3% / SME 6.3% and \"strained\" (OeNB, Apr 2026). Insolvencies \"risen sharply\" (ECB FSR, May 2026).\n\nCorporate insolvencies\n\nGermany: IWH monthly early indicator 57% above the pre-pandemic norm, May 2026 (1,518 cases); full-year 2025 ~23,900 (+8.3%), highest since 2014. Austria 6,810 (+3.4%), property failures +35% (Jan 2026). Italy ~+38%, among Europe's steepest (Oct 2025). IWH Insolvenztrend · Reuters/Destatis · KSV1870 · Allianz Trade.\n\nBanca d'Italia Financial Stability Report No.2 2025\n\nItalian small banks showed \"overall resilience\" in stress tests (Nov 2025). bancaditalia.it\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: ECB NPL data tables, Analysis hub, Quant apps hub\n\nMore posts: Bank Loss Rates Look Calm. The Stuff Underneath - Consumer, Private Credit, Non-Banks - Says Otherwise., ECB Stress-Tested Private Credit. Pensions Worst at 6% of Assets. Almost All the Damage Is the Wider Market Crash It Sets Off., Auto Losses Past the GFC Peak. Cards Almost There. Credit Unions Already at GFC Loss Rates. NY Fed + NCUA Data Now on Tigzig., The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top."
},
{
"slug": "ecb-stress-test-private-credit-may2026",
"title": "ECB Stress-Tested Private Credit. Pensions Worst at 6% of Assets. Almost All the Damage Is the Wider Market Crash It Sets Off.",
"tags": [
"portfolio-quants"
],
"content": "ECB Stress-Tested Private Credit. Pensions Worst at 6% of Assets. Almost All the Damage Is the Wider Market Crash It Sets Off.\n\nPublished: 2026-06-02\n\nThe ECB just ran a three-stage stress test on private credit, part of the May 2026 Financial Stability Review special feature. Stage 1: direct private credit losses. Stage 2: contagion to leveraged loans and high-yield bonds. Stage 3: tail scenario (equities -30%, HY repriced -25%, money runs out of private credit funds). Almost all the simulated damage is Stage 3 - the wider market crash the scare sets off, not the loans themselves. Severe-scenario losses: insurance corporations ~4% of assets, pension funds ~6% (the worst-hit), banks barely move. Euro area's direct private credit exposure is tiny (insurers 2.3%, banks 0.2%), but the US owns multiples more - insurers ~10% (15%+ for PE-owned), banks ~1.3-7%, with $300B in private credit and $1.65T in broader non-bank lending. Author commentary: contagion is bigger than the asset (as in prior crises), and given simultaneous global pressures plus the dot-com / GFC playbook, this should not be treated as 'just tail risk' (GFC 2008 was supposed to be tail risk too, until Lehman). ECB report: 'Stress in global private credit markets and its implications for euro area financial stability.'\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/ecb-stress-test-private-credit-may2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/ecb-stress-test-private-credit-may2026 with header Accept: text/markdown\n\nTags\nportfolio-quants\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Private Credit & NDFI Tracker, Analysis deck, Quant Apps hub\n\nMore posts: Private Credit and BDC Stress. Discounts at 13-Year Lows. Withdrawals Past the Gate. Regulators Now Arriving., FSB Just Published 50 Pages on Private Credit. The Title: 'Vulnerabilities in Private Credit.' Worth Reading., The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top."
},
{
"slug": "edgartools-sec-edgar-python-library",
"title": "Found a Python library that does all the heavy lifting for working with SEC EDGAR API - EdgarTools from Dwight Gunning",
"tags": [
"portfolio-analytics",
"fastapi"
],
"content": "Found a Python library that does all the heavy lifting for working with SEC EDGAR API - EdgarTools from Dwight Gunning\n\nEdgarTools\n\nHe has solved some really hard problems there.\n\nI am building a tool to compare quarterly financials across companies - Microsoft vs Google vs Amazon, last 12 quarters, custom views (side-by-side, QoQ, YoY, ratios and charts exactly the way I want it in terms of formats and computations).\n\nThe problem: SEC EDGAR has a great data API. But still a bit of pain and time-consuming to parse, standardize and organize things even with AI Coders (I use Claude Code).\n\nAsked Claude Code to find existing solutions. Don't reinvent the wheel.\n\nEdgarTools came up. Now using it as the backbone for a FastAPI backend - wrapping the library functions as API endpoints, React frontend pulls the data.\nXBRL standardization is built-in. You can compare Revenue, Net Income, Cash Flow across companies without manually mapping different tags.\n10-30x faster than alternatives. Uses PyArrow and lxml. Data comes back as pandas DataFrames.\nRespects SEC rate limits automatically - self-throttles to 9 requests per second. Proxy support if needed.\nCovers 10-K, 10-Q, 8-K filings, 13F fund holdings, Form 4 insider transactions, full financial statements.\nBuilt-in MCP server and AI Skills for Claude Code if you're doing SEC analysis with AI tools.\n\nI'm using this for my own quarterly analysis first. Once I get the format right, I'll publish the tool - open source like my other apps.\n\nIf you work with SEC filings and don't want to spend time on parsing and standardization, use this library. It does the heavy lifting.\n\nResources\nGitHub - EdgarTools\nEdgarTools Website\nDocumentation\n\n\nRelated\n\nTools: QRep - Security Reports, Quants Agent, Stock Data to AI Reports\n\nExplore: Quant apps hub, APIs and MCP, Build guides\n\nMore posts: QRep - Powered by QuantStats. Live Now., Security Performance Report for Investors. AI Quant Agent. Live. Open Source. Free., Architecture & Setup for a Dashboard with Hundreds of Millions of Records - Powered by DuckDB"
},
{
"slug": "enhancement-ai-technical-analysis-now-supports-multiple-llm-choices",
"title": "[ENHANCEMENT] AI Technical Analysis Now Supports Multiple LLM Choices",
"tags": [
"technical-analysis"
],
"content": "\\[ENHANCEMENT\\] AI Technical Analysis Now Supports Multiple LLM Choices\n\nGo to quants-suite.tigzig.com\n\nEnter Yahoo Finance symbol, choose time frames, pick your LLM.\n\nYou get a technical analysis report with daily and weekly charts (price, volume, technicals) with AI analysis and outlook - PDF and HTML format. I use this as a 2nd opinion against my own analysis.\n\nLLM choices\nFast tier (\\~30 seconds): Gemini 2.5 Flash/Lite\nMid tier (\\~1 min): GPT 4.1/Mini/Nano, Haiku 4.5\nPremium tier (2-5 mins): GPT 5.1, Claude Sonnet 4.5\n\nThe time-quality tradeoff: All models handle basic technical analysis well. Premium ones catch a few more patterns, offer slightly more nuanced analysis. Whether that's worth the extra time - your call.\n\nDocs and source code: Hit 'Docs' on the app site.\n\n\nRelated\n\nTools: Quants Suite, Quants Agent, QRep - Security Reports\n\nExplore: Quant Apps hub, Technical Analysis via MCP, Analysis hub\n\nMore posts: Quants Suite. 5 Reports. Performance, Risk & Technical Analytics., Quants Agent: Now with LLM Choices for Technical Analysis Reports, Nifty and S&P 500 Down - Run a Technical Analysis Check. 9 LLMs Compared., AI Technical Analysis Tool"
},
{
"slug": "evaluating-a-stock-pull-its-vigil-report",
"title": "Evaluating a stock? Pull its VIGIL Report. Free.",
"tags": [
"vigil",
"open-source"
],
"content": "Evaluating a stock? Pull its VIGIL Report. Free.\n\nVIGIL Corporate Signals Report\n\nSurveillance, insider, takeover, ratings, pledges, related party - the whole jing bang. One click.\n\nRaw data as filed. Nothing filtered, nothing summarized. All columns included with additional value-add fields.\n\nFive formats: Excel, JSON, CSV, TSV, SQLite.\n\nTSV and SQLite are particularly AI-friendly - paste TSV directly into context, or hand the SQLite file to your AI coder for large datasets. They love it.\n\nFree. Public data. No Login. No Ads. One click. Matter over.\n\nvigil.tigzig.com\n\n\nRelated\n\nTools: VIGIL - India Red Flag Events, QRep - Security Reports, Quants Agent\n\nExplore: Quant Apps, APIs and MCP, Browse all apps\n\nMore posts: VIGIL Data API and Download - 315K+ Records, 7 Tables, Updated Daily, New Tool Release - VIGIL: Credit Ratings, Pledges and Insider Trading for India Markets, Related Party Transactions - Now Live on VIGIL, New on VIGIL: SAST Takeover Disclosures (India)"
},
{
"slug": "execute-asap-approval-granted-google-vs-microsoft-meta",
"title": "Execute ASAP. Approval granted. Google - against Microsoft & Meta past 15 years, benchmark vs. S&P 500, technicals & quarterlies for all three.",
"tags": [
"portfolio-analytics",
"technical-analysis"
],
"content": "Execute ASAP. Approval granted. Google - against Microsoft & Meta ...\n\n\"Execute ASAP. Approval granted. Google - against Microsoft & Meta past 15 years, benchmark vs. S&P 500, technicals & quarterlies for all three.\"\n\nThat's the prompt.\n\nPass it on to my Quants Agent at quants.tigzig.com\n\nGet the full report pack: PDF, HTML and CSV for offline analysis.\n\nBe patient. Takes around 2 mins to generate the full set.\n\nWhat you get\n\nSecurity Performance Report (CAGR, Sharpe, Sortino, Drawdowns, Monthly Returns), AI Technical Analysis, QuantStats Report, Financials and Price Data (CSV/Excel)\n\nCan the quants agent analyze the reports too?\n\nNot yet. Not trained it. But go ahead and ask. Sometimes has a mind of its own.\n\nHow does quant agent work?\n\nTriggers API call to FastAPI endpoints where the heavy lifting happens and reports get created.\n\nWill it customize reports and UI?\n\nNot its job. That's what AI coders are for - Claude Code and Cursor. They'll do frontend, backend and everything in between. If you want guides on working with AI coders, check my blog at tigzig.com.\n\nEnjoy your Sunday. I'm off cycling. But my 30-member team is working hard and round the clock at www.tigzig.com - ready to beat and torture data until it confesses whatever you wish.\n\n\nRelated\n\nTools: Quants Agent, QRep - Security Reports, Quants Suite\n\nExplore: Quant Apps hub, API/MCP hub, Analysis hub\n\nMore posts: Security Performance Report for Investors. AI Quant Agent. Live. Open Source. Free., Quants Agent: Now with LLM Choices for Technical Analysis Reports, Three Tools I Built for Market Analysis - QSUITE, TREMOR, QREP"
},
{
"slug": "extract-python-code-from-xlwings-lite-excel-files",
"title": "How to Extract Python Code from xlwings Lite Excel Files",
"tags": [
"xlwings-lite",
"converters-tools"
],
"content": "How to Extract Python Code from xlwings Lite Excel Files\n\nYou can do this yourself or just copy paste the full post to your AI Coder (Claude Code , Cursor, Gemini etc) and it will do it for you in seconds.\n\nEdit: Felix Zumstein pointed out a simpler way which I had missed - just click \"Don't Trust\" when the warning pops up and copy the code from the editor. Matter over. If you still want to extract the code without opening Excel at all, or want to understand where xlwings Lite stores its files, read on.\n\nPurpose\n\nExtract main.py and requirements.txt from an xlwings Lite .xlsx file without opening it in Excel. Useful for inspecting code before opening untrusted files, or for extracting scripts for version control.\n\nHow it works\n.xlsx files are ZIP archives containing XML files\nxlwings Lite stores Python code in: xl/webextensions/webextension1.xml\nThe code is in a \n\n
\n\nCORS - Cross-Origin Resource Sharing\n\nIf you are building browser tools, AI or otherwise, you have run into CORS issues. If not, you will. Maybe you tried to pull Yahoo Finance data straight into a browser app, or download a file from GitHub or Google Drive into your web tool, and the browser just blocked it.\n\nThis post does two things. First, it explains CORS in plain words - what it is, why it blocks you, and the traps to avoid. Second, it gives you a complete, free fix: a copy-paste Cloudflare Worker that works as a CORS proxy - no server, no custom domain, free tier. The full Worker code is at the bottom of the post, with a one-line tweak to point it at Yahoo Finance or any API you need.\n\nWhat is CORS?\n\nWhen you download from a terminal (like using curl or python) - this always works. No restrictions.\n\nBut inside a browser (JavaScript code running in a web app) - there are restrictions. When your browser code tries to fetch data from another website (like Google Drive), that server must explicitly say \"Yes, browsers are allowed to access this.\" They do this by including a special header in the response: Access-Control-Allow-Origin: *\n\nThis is CORS - Cross-Origin Resource Sharing. It's a browser security feature.\n\nWhat's the problem?\n\nMany services (e.g. Yahoo Finance, GitHub Releases, Google Drive, Dropbox etc) don't include this header. So your browser blocks your JavaScript from reading the response. It says: \"Sorry, this server didn't give me permission to share this data with a browser.\"\n\nWait - does the data actually reach the browser?\n\nYes. This is the confusing part. The browser makes the request, the server sends the data back. The data actually arrives at your browser. But before handing it to your JavaScript code, the browser checks for that Access-Control-Allow-Origin header. If it's missing - your JavaScript can't touch the data. The browser's security layer sits between the network and your code.\n\nLooking for a free public CORS proxy URL you can just paste in and use? Read this first.
\n\nQ: Do you run a free public CORS proxy I can point my app at?
\nNo, I do not. And honestly, you should be careful about using anyone else's either. Let me explain why in plain terms, then show you the better way. It is free and takes about 5 minutes.
Q: Why not just use one of the free public ones?
\nA public CORS proxy is a stranger's server that every one of your requests passes through. Three real problems:
- \n
- They can see everything. Every URL you call, and anything inside it - API keys, tokens, query parameters - goes through their machine. They can log it, read it, even change what comes back to you. \n
- They disappear. The free public ones get overloaded, rate-limited, and shut down all the time. Build your app on one and it breaks the day that proxy goes away. \n
- You inherit their bad reputation. If other people abuse that shared proxy, the API you are calling can block its address - and now your calls fail too, through no fault of your own. \n
Q: So what should I do instead?
\nRun your own tiny copy. It is free, it takes about 5 minutes on Cloudflare, and then the proxy is yours - no stranger in the middle, no surprise shutdown. This post explains CORS in plain words and gives you a simple GET-only file proxy (for GitHub, Google Drive, Dropbox, Yahoo Finance and the like). If you need a general one for any URL and any method, including POST, use the CORS Anywhere alternative. Full copy-paste code is in both - you can hand it straight to your AI coder.
\nNuance: Sometimes the data doesn't even arrive - Preflight Requests\n\nEverything above applies to simple requests - basic GET requests that just read data. But for more complex requests - POST with JSON body, requests with custom headers like
\n\nSo where is the data sitting?\n\nIt's in the browser's network layer. The bytes came in, the HTTP response was parsed. But it's walled off from your JavaScript. Think of it like a customs officer who received your package but won't hand it to you because the paperwork (CORS headers) is missing.\n\nCan I see it in DevTools?\n\nYes! This trips people up. Go to DevTools > Network tab, click the request - you can see the full response body with the actual data right there. You can read it. The browser has it.\n\nBut your fetch() call still throws a CORS error. DevTools has elevated privileges that bypass the same-origin policy. So you're staring at the data in DevTools while your code can't access it. Frustrating.\n\nWhat's the solution?\n\nYou add a middleman - also called a Proxy. Instead of fetching directly from the offending API, your browser fetches from a proxy. The proxy fetches the data server-side (which always works - CORS is a browser-only restriction), then sends it to your browser with the proper CORS headers added. Your browser is happy, and you get your data.\n\nCORS issues are fairly common when building browser-based apps. Knowing how to use a proxy to bypass CORS is an essential skill if you're building deployable apps and data tools.\n\nAuthorization, PUT, DELETE - the browser does something different.\n\nWhat is a preflight request? Before sending the actual request, the browser sends a lightweight OPTIONS request first. It asks the server: \"Would you accept this kind of request from a browser?\" If the server says yes (returns the right CORS headers), the browser sends the real request. If not, it stops right there. The real request never goes out. The data never arrives.\n\nWhy does the browser send a preflight? It's a safety check. A POST or DELETE can change data on the server - create records, delete data, trigger actions. The browser's logic: \"Before I send something that could modify data on a server that might not be expecting browser requests, let me check first.\" This protects older APIs that were never designed to be called from random websites.\n\nCan I disable it? No. The browser decides automatically. You cannot tell the browser \"skip the preflight.\" It's not a setting or a flag in your code.\n\nHow does the browser decide when to send a preflight? Simple rules. The browser skips preflight (sends directly) if ALL of these are true: method is GET, HEAD, or POST; only standard headers (no Authorization, no custom headers); and if POST, content type is only form data or plain text. The moment you break any of these - say you add Content-Type: application/json or an Authorization header - the preflight happens automatically. Most modern API calls use JSON and auth headers, so most calls trigger preflight.\n\nDoes it slow things down? It's an extra round-trip but it's fast (just headers, no body) - you generally don't notice it.\n\nThe fix is the same either way - use a proxy. But now you know why sometimes you see that OPTIONS request in your DevTools Network tab before the actual call.\n\nTrap: Don't use mode: 'no-cors' - it doesn't fix anything\n\nWhen you hit a CORS error, you'll find suggestions online to add
\n\nHow I handle it - and which one to pick when\n\nNumerous ways to set up a proxy. I typically end up using one of these 3. Here's the order I think about them.\nCloudflare Workers - The Pure Proxy (Free. No Domain Required)\n\nThis is the simplest option for pure pass-through. Your browser calls the Cloudflare Worker, the Worker fetches from the actual API, adds CORS headers, sends it back. That's it. No domain needed - you get a free workers.dev subdomain. No server to manage. Free Cloudflare account is enough. You can also add authentication if needed.\n\nGreat for: any situation where you just need to pass data through without processing it. If the API you're calling works fine but just doesn't have CORS headers - this is the quickest fix. The complete copy-paste Worker code is at the bottom of this post - jump to The Complete Cloudflare Worker CORS Proxy. No second page needed. There is also a standalone text guide in the xlwings Lite Data Importer you can copy straight to your AI coder.\nVercel Serverless Functions - The Mini Backend (Free Tier)\n\nThis is a very nifty feature. If you have a React app on Vercel, you can create an api/ folder in your project. Any file inside api/ becomes a serverless function - essentially a mini backend endpoint. It runs on Vercel's servers, not in the browser. So no CORS issues because the request goes from Vercel's server (server-to-server), not from the browser.\n\nIt's like getting a mini FastAPI backend for free - except in JavaScript. You can do fair amount of processing there - not just proxy calls. Anything that can be done in JavaScript, you can do there. Data transformation, API orchestration, auth token handling.\n\nThe free tier gives you up to 12 serverless functions - more than enough for most apps. If you run into the limit, you can always combine functions. Execution timeout on free tier is currently up to 300 seconds (5 minutes) - so even long-running queries or heavier processing work fine.\n\nFor local development, you need the Vercel CLI (vercel dev) to run serverless functions locally - they won't work with just npm start since they need Vercel's runtime.\n\nIn my newer React apps, I've moved all API calls from the frontend to serverless functions. Everything goes through the server side. This adds an extra layer of security (API keys never exposed in browser code) and CORS issues are automatically taken care of. That's my default approach now.\n\nExample: My IMDb Movie Explorer app. For source code hit 'Docs' on the app site (Press F9 if you don't see top menu).\nFastAPI Backend - For Heavy Python Processing (Not Free)\n\nThis is not required for all cases. For simple pass-through requests, this is overkill - Cloudflare Workers or Vercel serverless handle that easily.\n\nWhere this makes sense: when you need serious Python processing that can't be done in JavaScript. Libraries like yfinance, pandas, numpy, scikit-learn - things that only exist in the Python ecosystem. Or when you already have a Python backend doing other things and you just want to add another endpoint.\n\nExample: Yahoo Finance Data Extractor. A FastAPI backend that runs yfinance. Public and open. Pull Yahoo Finance data from any browser.\n\nSo which one do I pick?\n\nTypically: if I'm building a React app, Vercel serverless functions are my first choice by default - everything goes through the server side. If something can't be done there (needs Python), I use my FastAPI backend. For standalone pass-through where I just need to proxy an API call quickly, Cloudflare Workers - doesn't even need a domain.\n\nmode: 'no-cors' to your fetch call. It looks like a fix. The error goes away. But your data is gone too.\n\nWhat no-cors actually does: it tells the browser \"I know this will fail CORS, send it anyway but I accept that I can't read the response.\" The browser makes the request, gets the response, and gives you an empty opaque response. No data. No error either. Just nothing.\n\nIt's useful in very specific cases (like sending analytics pings where you don't care about the response), but for fetching data - it's a trap. The CORS error disappears but so does your data. Use a proxy instead.\n\nlocalhost:3000 calling localhost:8000 is also cross-origin\n\nThis catches people during development. Your React app runs on localhost:3000. Your backend runs on localhost:8000. Same machine. Same \"localhost.\" But different ports. To the browser, these are different origins. So yes - you get CORS errors even when both are running on your own laptop.\n\nThat's why your FastAPI backend needs CORS middleware even during local development. In FastAPI you add
\n\nThe Complete Cloudflare Worker CORS Proxy (Copy-Paste Code)\n\nHere is the whole thing. This is the exact Worker I use. It is free, needs no custom domain (you get a your-name.workers.dev URL automatically), no server to manage, and it streams files of any size. You can have it live in a few minutes - or just paste this whole section to your AI coder and let it deploy for you.\n\nHow you call it\n\nOnce it is deployed, you call it by putting your target URL in a ?url= parameter (URL-encoded):\n\n``\nhttps://your-worker.workers.dev/?url=CORSMiddleware with allow_origins=[\"*\"] (or specific origins for production). Without it, your local frontend can't talk to your local backend. Same machine, different ports, different origins.\n\n
\n\nFree to use. If it helps your work, a link back to tigzig.com is appreciated.\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, Quants Agent, QRep - Security Reports\n\nExplore: Mutual Fund hub, Market tools hub, Claude for Analytics\n\nMore posts: Now live - MFPRO v2 - Mutual Fund Analytics (India) - now with rolling returns, custom eval periods, multi-period & multi-instrument comparisons, Rolling Returns: Why CAGR Alone Can Mislead You (And What To Use Instead), Analysis-as-App: Inside India’s Top Midcap Funds: Buys, Sells, Entries and Exits. Interactive Dashboard Release (Analysis-as-App)"
},
{
"slug": "free-production-grade-databases-get-setup-in-minutes-great-for-testing-and-development",
"title": "Free, Production-Grade Databases. Get setup in minutes. Great for testing and development",
"tags": [
"database-ai",
"infrastructure"
],
"content": "Free, Production-Grade Databases. Get setup in minutes. Great for testing and development\n\nNeon: My top choice for AI apps. Database live in <1 sec. Straightforward API for creating / modifying databases and role.\n\nSupabase: Auth integration.\n\nAiven: Most generous free tier (5GB) and supports both Postgres & MySQL.\n\nI use these for creating a 'one-time-use' database for testing to live client projects.\n\nSee live in action on www.tigzig.com -> Database AI & SQL Apps\nDATS-4: temporary on-the-fly databases powered by Neon\nCustom GPT: powered by Supabase with Cricket ODI data\nRealtime Voice AI: powered by Aiven\n\nAll open source. Hit 'Docs' on app site for source code\n\nThe attached image is a quick-glance spec sheet comparing my notes on each\n\nFree Database Providers Comparison\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, Multi-Database GPT, Voice DB - ElevenLabs\n\nExplore: Database AI Apps hub, API/MCP hub, Browse all apps\n\nMore posts: Instant Database Setup for AI Apps. With Neon.com, Connect ChatGPT to Supabase in 10 mins., Cricket (ODI/T20) & Tour de France stats from a Custom GPT connected to 3 live databases."
},
{
"slug": "from-12-second-queries-to-under-1s-optimizing-230-million-row-dashboard",
"title": "From 12 second queries to under 1s: Optimizing a 230 Million Row Dashboard - 14 Bottlenecks I Had to Fix",
"tags": [
"duckdb",
"fastapi",
"infrastructure"
],
"content": "From 12 second queries to under 1s: Optimizing a 230 Million Row Dashboard - 14 Bottlenecks I Had to Fix\n\nCine Pro Timeline\n\nThe first version was taking 9 to 12 seconds. I would click, wait, wonder if something was broken. The database had 230 million rows across multiple tables. The largest single table: 91 million records. Total size: 16GB.\n\nThis is the story of getting that down to under 3 seconds - most pages under a second.\n\nI built this with Claude Code. The process was methodical: add a timer to each component, measure page load box by box, then tackle the slowest one. Claude Code would run investigations - writing test scripts, timing queries with curl and Python, tracing execution paths through frontend and backend. It would come back with a diagnosis: \"This query is scanning 12 million rows because of a JOIN. Here's why the index isn't being used.\"\n\nOne problem at a time. Measure, diagnose, fix, measure again.\n\nThe app now runs on two backends simultaneously - Hetzner in Germany and Oracle Cloud in USA - with a frontend toggle to switch between them. Users can compare latency from different regions. Both backends serve the same 16GB DuckDB database with identical optimization.\n\nHere's everything I learned getting from 9 to 12 seconds to 0.5 to 2.5 seconds.\n\nSection 1: Pre-Compute Everything\n\nThe single biggest performance lever. Four separate bottlenecks, all solved by the same idea: do the expensive work once... create flat tables with pre-aggregations, add index, query the result.\n\nChallenge 1: The JOIN That killed performance\n\nThe Deep Profile page shows a person's complete filmography - every movie, TV show, and short they've worked on. For someone like Morgan Freeman, that's 200+ titles with ratings, vote counts, and genres.\n\nThe original query looked reasonable:\n\n``sql\nSELECT tb.primaryTitle, tb.startYear, tr.averageRating\nFROM title_principals tp\nJOIN title_basics tb ON tp.tconst = tb.tconst\nLEFT JOIN title_ratings tr ON tb.tconst = tr.tconst\nWHERE tp.nconst = 'nm0000151'\n`\n\nTime: 1.6 seconds. Every single time.\n\nThe problem: DuckDB wasn't using the index on titlebasics for the JOIN. It was doing a full sequential scan of 12 million rows to find the ~200 matching titles. The filter on titleprincipals was fast (indexed), but the JOIN to title_basics killed performance.\n\nThe fix: Pre-compute the JOIN once, store it as a denormalized table:\n\n`sql\nCREATE TABLE person_filmography AS\nSELECT tp.nconst, tp.tconst, tp.category,\n tb.primaryTitle, tb.startYear, tb.titleType,\n tr.averageRating, tr.numVotes\nFROM title_principals tp\nJOIN title_basics tb ON tp.tconst = tb.tconst\nLEFT JOIN title_ratings tr ON tb.tconst = tr.tconst\n`\n\nResult: 91 million rows, ~1.8GB storage. But now every person lookup is a simple indexed filter - no JOINs at query time.\n\nBefore: 1.6 seconds (JOIN with full table scan)\nAfter: 0.5 seconds (indexed filter on denormalized table)\nImprovement: 3.4x faster\n\nThe principle: if you're repeatedly JOINing the same tables for the same access pattern, just create the JOIN once and push into a intermediate table\n\nChallenge 2: 15 HTTP Requests for a Static Page\n\nThe main dashboard shows database statistics, rating distributions, top movies by genre, hidden gems. None of this data changes - it's aggregate stats from a dataset that updates monthly at most.\n\nBut the page was making 15 parallel API calls on every load. Each request: DNS lookup, TLS handshake, routing through Vercel serverless, hitting the backend, returning JSON. Even with backend caching, the network overhead alone was 2-3 seconds.\n\nI watched the Network tab: 15 requests, 15 responses, all returning small JSON payloads. Total data: maybe 100 rows across all queries. Total time: 3 seconds.\n\nThe page is static. Every user sees the same thing. Why query 15 times?\n\nFix: Store the entire dashboard response as a single JSON blob in DuckDB. One table, one row, one query returns everything. The JSON gets regenerated when the data pipeline runs (weekly).\n\nBefore: 15 HTTP requests, ~3 seconds\nAfter: 1 HTTP request, ~0.65 seconds\nImprovement: 4.6x faster\n\nChallenge 3: In-Memory Query Cache - Why Most Queries Never Hit DuckDB\n\nAll the pre-computation above makes individual queries fast. But the fastest query is the one you don't run at all.\n\nThe backend has an in-memory cache sitting in front of DuckDB using cachetools (Python library). Every query result gets cached by its SQL hash. Same query comes in again - return the cached result, skip DuckDB entirely.\n\nMost dashboard usage is repetitive. People search for popular actors. The same filmography queries get run over and over. After the first user looks up Morgan Freeman, every subsequent lookup is a cache hit - zero DuckDB work.\n\nCache resets on server restart since it's in-memory, not persisted. Cold starts are slower, but the cache warms up quickly with normal traffic. Config is 2000 entries max per database file with a 30-day TTL. Enough to cover the popular queries without eating too much RAM.\n\nChallenge 4: Same Principle, Applied Three More Times\n\nOnce I saw pre-computation working, I applied it everywhere the pattern fit:\n\nLeaderboards: The \"Most Prolific\" section shows top actors, actresses, and directors by era - six eras, three categories. Each combination was scanning 97 million rows, taking 2-5 seconds per query. Pre-computed all combinations into a summary table.\nBefore: 2-5 seconds. After: ~50ms. Improvement: 40-100x faster.\n\nPerson stats: The stats header (total titles, average rating, career span) was running 6 parallel aggregation queries per person, adding up to ~300ms. Created a person_stats table with one row per person.\nBefore: ~300ms. After: ~5ms. Improvement: 60x faster.\n\nSearch ranking: Search for \"fost\" and Adam Foster appeared before Jodie Foster - alphabetical order, ignoring that Jodie has 700+ credits. Used the pre-computed personstats table to order results by totaltitles. Famous people show up first. And because the stats are pre-computed, the ORDER BY adds ~0.1ms per result instead of ~5-10ms with runtime aggregation.\n\nStorage is cheap. User patience is not.\n\nSection 2: Query Patterns That Backfire\n\nChallenge 5: Searching 15 Million People for Every Keystroke\n\nThe person search queries name_basics - 15 million people. Most are extras, background actors, or people with a single credit. Type \"John\" and the database scans all 15 million rows. About 2 seconds per search.\n\nBut most users search for known actors and directors. Out of 15 million people, only 2.3 million have 3 or more credits. That smaller set covers nearly all searches anyone will actually run.\n\nFix: Created a prolific_people table - just people with 3+ credits. Added a \"Smart Search\" toggle, on by default. Searches scan 2.3 million rows instead of 15 million.\n\nBefore: ~2 seconds (scanning 15M rows)\nAfter: ~200ms (scanning 2.3M rows, 85% less data)\n\nIf someone doesn't show up, uncheck Smart Search to search the full database. In practice, nearly all searches are covered by the smaller set.\n\nChallenge 6: The ORDER BY That Killed My Index\n\nAfter all the pre-computation work, one query was still slow: the filmography list. EXPLAIN ANALYZE showed the problem:\n\n`\nTABLE_SCAN - Type: Sequential Scan\nRows scanned: 91,479,013\n`\n\n91 million rows scanned. But I had an index on nconst. Why wasn't DuckDB using it?\n\nThe query:\n\n`sql\nSELECT * FROM person_filmography\nWHERE nconst = 'nm0000151'\nORDER BY startYear DESC\nLIMIT 500\n`\n\nThe problem: DuckDB's ART indexes have a critical limitation. They only work for single-column equality/IN conditions. The moment you add ORDER BY on a different column, the optimizer decides a full table scan is \"cheaper\" than index lookup + sort. Remove the ORDER BY, and the EXPLAIN output changes to Index Scan.\n\nThe fix: Remove ORDER BY from SQL, sort in JavaScript instead.\n\n`javascript\nfilmography.sort((a, b) => (b.year || 0) - (a.year || 0))\n`\n\nSorting 500 objects in JavaScript: less than 1ms. Saved by removing ORDER BY: 1100ms.\n\nBefore: 2.0 seconds (Sequential Scan because of ORDER BY)\nAfter: 0.9 seconds (Index Scan, client-side sort)\nImprovement: 55% faster\n\nTakeaway: When DuckDB isn't using your index, check for ORDER BY. Remove it and sort client-side if the result set is small.\n\nChallenge 7: Adaptive Query - Check the Count Before You Query\n\nChallenge 6 removed ORDER BY to get Index Scan. But that created a new problem.\n\nLIMIT without ORDER BY doesn't return the first N or the last N rows. It returns any N - whatever DuckDB finds first in storage order. For someone with 200 titles and a LIMIT of 5000, we get all 200. Client-side sort works fine. But for someone with 39,000 titles, LIMIT 5000 returns an arbitrary 5000. Client-side sort then orders this random sample by year. Completely wrong data.\n\nThe fix is an adaptive query. Before running the filmography query, make a separate API call to check how many titles this person has. One row from person_stats. The database lookup is instant, but this is a full round-trip through the network (frontend to Vercel to backend and back), so it costs a few hundred milliseconds.\n\nIf they have 5000 or fewer titles - download everything. No ORDER BY needed because LIMIT doesn't cut anything off. This is the fast path: Index Scan, ~0.9 seconds.\n\nIf they have more than 5000 titles - add ORDER BY to get the most recent 5000. This forces the slower Sequential Scan (~2 seconds), but we need the right data, not random data.\n\n99.98% of people have 5000 titles or less - they get the fast path. Only 513 people in the entire database have more than 5000. The most prolific is Ekta Kapoor with 39,000 entries. The count check adds a network round-trip, but it saves over a second on the main query by keeping Index Scan for nearly everyone.\n\nAn extra API call to check the count costs a few hundred ms, but saves ~4.5 seconds on the filmography query for 99.98% of lookups.\n\nChallenge 8: EXISTS vs CTE - A 15x Performance Gap\n\nThe Discover page lets users filter movies by person - \"show all Morgan Freeman movies.\" The original query used EXISTS:\n\n`sql\nSELECT tb., tr.\nFROM title_basics tb\nJOIN title_ratings tr ON tb.tconst = tr.tconst\nWHERE EXISTS (\n SELECT 1 FROM title_principals tp\n WHERE tp.tconst = tb.tconst AND tp.nconst = 'nm0000151'\n)\n`\n\nTime: 2-3 seconds. The EXISTS subquery runs for every row being evaluated - millions of index lookups.\n\nThe fix: CTE pattern. Get the person's titles first (fast indexed lookup), then JOIN to that small set:\n\n`sql\nWITH person_titles AS (\n SELECT DISTINCT tconst\n FROM person_filmography\n WHERE nconst = 'nm0000151'\n)\nSELECT tb., tr.\nFROM person_titles pt\nJOIN title_basics tb ON pt.tconst = tb.tconst\nJOIN title_ratings tr ON tb.tconst = tr.tconst\n`\n\nThe CTE returns ~200 titles. The main query joins to this tiny set instead of running EXISTS against millions of rows.\n\nBefore: 2-3 seconds (EXISTS subquery on every row)\nAfter: ~200ms (CTE + small join)\nImprovement: 10-15x faster\n\nTakeaway: When filtering by a related entity, get that entity's IDs first with a CTE, then JOIN to the small result set.\n\nChallenge 9: The \"Check First\" Anti-Pattern\n\nThe stats section needed to handle two cases: prolific people (in person_filmography table) and non-prolific people (fallback to slower JOINs). The original code ran a check query first - \"is this person in the fast table?\" - then ran the actual stats queries based on the result.\n\nThe filmography appeared instantly. Stats appeared 300ms later. The check query had to complete before the stats queries could start. Sequential bottleneck.\n\nFix: \"Try fast, fallback if empty\" pattern. Run all the fast-table queries immediately in parallel. If they come back empty, the person isn't in the fast table - run the fallback queries. For the 95% of lookups that hit the fast table, stats load instantly alongside the filmography. For the 5% that don't, there's a small penalty - fast queries return empty quickly, then fallback queries run.\n\nBefore: Check query (300ms) then stats queries (parallel)\nAfter: All queries (parallel), fallback if needed\nResult: Stats appear with filmography, not after\n\nTakeaway: Don't check which path to take. Try the fast path. Fallback if it fails.\n\nSection 3: Stop Querying What You Already Have\n\nChallenge 10: When 5 Database Queries Became Zero\n\nThe Top Stats Band shows highest rated titles, most voted titles, titles per decade, and title types. Original implementation: 5 parallel database queries, ~3-4 seconds total.\n\nThen I realized: the filmography tab loads up to 5000 titles for this person. Those 5000 rows contain everything needed for the stats - ratings, votes, years, title types. The data is already in the browser.\n\nFix: Compute stats from the filmography array in JavaScript. Array.sort on 5000 objects takes less than 5ms. The network round-trip it replaced took 3-4 seconds.\n\nBefore: 5 database queries, 3-4 seconds\nAfter: Array operations, less than 5ms\nImprovement: ~99% faster\n\nI applied this same pattern across the app. The filmography table has Role and Title Type filters - originally each filter change triggered a new database query (1-2 second wait). Now it's client-side Array.filter on the already-loaded data. Instant. The Compare tab showed side-by-side stats for multiple people - originally 5 database queries when you switched tabs. Now it computes comparison stats from the loaded filmography for the current person. Instant.\n\nThe principle: If you've already loaded the source data into the browser, don't make a round-trip to the database for derived views of that same data. Compute them client-side. For datasets under ~10,000 rows, client-side filtering and aggregation is almost always faster than a server round-trip.\n\nSection 4: Infrastructure Realities\n\nChallenge 11: The Container That Thought It Had 24GB of RAM\n\nOracle Cloud backend was 4x slower than Hetzner. Same code, same database, similar specs. OCI queries: ~4 seconds. Hetzner queries: ~1 second.\n\nInvestigated disk latency (ioping showed OCI was 10x slower - red herring, DuckDB caches in RAM). Investigated network (similar). Finally checked memory:\n\n`bash\nContainer limit\ndocker inspect Prefer a screen to code? Analyze it in MFPRO.
\nMFPRO runs on this same AMFI data, in your browser. No key, no download, no code. Pick up to 200 funds (or start from a ready-made preset of 24 or 104), add any Nifty or BSE benchmark, and compare over any period.
\nReturns: CAGR across windows, Sharpe, Sortino, alpha, beta, R-squared, up and down capture, drawdowns and Calmar, VaR, and rolling returns (average, median, min, max, beat %). 40+ return, risk and drawdown metrics in all.
\nComposition: for 24 major equity funds (about a third of the equity universe), drill into holdings, entries and exits, and index composition.
\nOpen MFPRO →\n![\"MFPRO](\"/images/blog/mfprodatadictionary_onepager.png\")
\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, Quants Agent\n\nExplore: Mutual fund analytics hub, MF NAV API docs, Full API + MCP catalog\n\nMore posts: One Line of User Feedback, Fixed Within a Day. MFPRO Search Upgraded + Complete AMFI NAV History, 37M Rows, 2006 Onward., MFPRO Data & API Screen Live - Download the Whole India MF NAV Database in One Click. No Code, No API., Free Mutual Fund NAV API for India - Revamped. AI Ready. 18,000+ Schemes Since 2008 from AMFI., Rolling Returns: Why CAGR Alone Can Mislead You (And What To Use Instead)"
},
{
"slug": "mfpro-download-truncation-fix-jul2026",
"title": "1 in 5 Bulk Downloads Were Silently Truncating. Two Cloudflare Worker Bugs, Found and Fixed. Now Up to 4x Faster.",
"tags": [
"infrastructure",
"mutual-funds"
],
"content": "1 in 5 Bulk Downloads Were Silently Truncating. Two Cloudflare Worker Bugs, Found and Fixed. Now Up to 4x Faster.\n\n1 in 5 bulk downloads of the Tigzig MFPRO India mutual fund database were quietly running into trouble .. silent truncation, retries, higher latency. Now fixed. And up to 4x faster.\n\nMy dashboard tracks every error Cloudflare sees at the edge, including the class of failures my own server logs can never see. A few days back a pattern surfaced there, 504s on the full-database downloads, roughly 1 in 5 attempts. There was no error page for the user. The download would start fine, run for a while, then the file would arrive incomplete, and users on slower connections were hit hardest.\n\nI dug in with Claude Code, reproduced it with throttled downloads from multiple countries, and found two separate bugs in how my Cloudflare Worker was serving large files from R2 storage.\n\nLessons for fellow tool builders, especially if you are serving large files\nDo not feed a user's download stream into your edge cache write. Fill the cache with a separate read, otherwise the cache write can finish first and cut the user off mid-file.\nDo not pass a large file through JavaScript inside a Worker. Stream it native, or the CPU limit will truncate it silently.\nMore than one way to do this, and situations differ. But whatever you do, run test scenarios, and most importantly, monitor all API calls, edge and origin both, errors, latencies and bytes served, to catch truncations.\n\nAnd be assured, however much you test, out in the wild new things will pop up .. your monitoring needs to be strong enough to catch those.\n\nBoth fixed now. Verified byte by byte from four countries. Downloads now arrive complete, and up to 4x faster.\n\nMFPRO India Mutual Fund API - Docs Hub: tigzig.com/mfpro -> Docs\n\nWhat is actually in the full database download?\n\nOne file, every NAV AMFI has ever published. 37M+ rows across 38,000+ schemes, 2006 onward, matured and dead schemes included. Five columns: scheme code, date, NAV, scheme name, ISIN.\n\nSeven file options to pick from: Parquet 167 MB, CSV or TSV compressed at about 236 MB each, SQLite 1.28 GB. All seven carry the identical 37,084,484 rows (as of now), regenerated two to three times a day from AMFI. A manifest endpoint publishes the current sizes, row count and a validation flag, so your script or agent can confirm it got the whole thing.\n\nOne GET, no key: api.tigzig.com/mf/v1/download?format=parquet\n\nParquet loads into pandas in one line.\n\nOr skip the download entirely, DuckDB can query the parquet straight over HTTP and pull only the byte ranges your query touches.\n\nDocs: tigzig.com -> MFPRO -> Docs\n\nMore where this came from\n\nThe MF NAV API is one of many APIs and MCP servers on tigzig.com, all no key, no sign up.\n\nTremor for global macro, 300+ indicators across US, India, Europe. Vigil for India red flags, filings, insider trades, pledges, ratings. Yahoo Finance extraction, security comparison, technical analysis reports, and a markdown to PDF engine. Almost every one runs as both a REST API and an MCP server on the same backend, so your Claude, Cowork or any agent can connect directly.\n\nLive uptime for the data APIs sits at status.tigzig.com.\n\nThe full directory: tigzig.com -> API/MCP.\n\nAnd if browsing is not your thing, just ask your AI to read api.tigzig.com and find what you need.\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, TREMOR - Macro Stress Signals, VIGIL - India Red Flag Events\n\nExplore: Mutual fund analytics hub, API + MCP hub, MF NAV API docs\n\nMore posts: 227 Downloads, 41GB in 10 Days. India MF Complete NAV Database - 38K+ Schemes, 37M+ Records, One 167MB Parquet. Free., Tigzig APIs - Reading the Logs, Shipping the Fixes. Status Page Live, R2 Downloads, Self-Correcting 4xxs., Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users, The boring $5 tool that tells me when a server goes down"
},
{
"slug": "mfpro-downloads-full-database-jul2026",
"title": "227 Downloads, 41GB in 10 Days. India MF Complete NAV Database - 38K+ Schemes, 37M+ Records, One 167MB Parquet. Free.",
"tags": [
"mutual-funds"
],
"content": "227 Downloads, 41GB in 10 Days. India MF Complete NAV Database - 38K+ Schemes, 37M+ Records, One 167MB Parquet. Free.\n\nIn the past 10 days, Tigzig MFPRO has served 227 downloads ... 41GB total ... of India Mutual Fund Complete NAV Database ... high speed from Cloudflare's global edge network via R2. Covering 38K+ schemes, 37M+ records on a 167MB Parquet file.\n\nIncludes closed schemes also .. so you can run a regime-based analysis going back more than two decades ... say you'd like to see how funds did during the 2008 Great Financial Crisis as well as other shock events post that.\n\nThe database is completely free, open, and gets refreshed three times every single day so you are always working with the latest NAVs.\n\nFreely downloadable ... manually from a browser as well as via API.\n\nJust head over to tigzig.com -> MFPRO -> Data & API\n\nOr just point your AI Agent at api.tigzig.com and have it download it for you.\n\nThe update process is run through a multi-step validation & comes with a data dictionary, build process, a field guide dissecting various scheme types as well as data anomalies seen. All available on the Docs section.\n\nThe scheme universe behind the file\n\nThe scheme codes include every plan and option variant. Currently NAV's for ~8,500 codes is published by AMFI daily, that's roughly 1,950 funds as AMFI counts them.\n\nOf the 38,000+ scheme codes that have ever published a NAV since 2006, 22% are active and rest are inactive, but very rich source for historical and regime based analysis.\n\nYou can run these cuts yourself on the interface via an interactive cross-tab feature: tigzig.com -> MFPRO App -> Explorer\n\nUptime\n\nAnd after some server issues earlier in the month, since the past ~15 days, the MFPRO API uptime is at 99.7% .. status available live here: status.tigzig.com\n\n![\"MFPRO](\"/images/blog/mfprodownloadsonepager.png\")
\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, Quants Agent\n\nExplore: Mutual fund analytics hub, MF NAV API docs, Full API + MCP catalog\n\nMore posts: Data Dictionary & Docs Live on MFPRO. Every Field Explained, 21-Metric Methodology, and How the India MF Universe Actually Looks., One Line of User Feedback, Fixed Within a Day. MFPRO Search Upgraded + Complete AMFI NAV History, 37M Rows, 2006 Onward., MFPRO Data & API Screen Live - Download the Whole India MF NAV Database in One Click. No Code, No API., Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users"
},
{
"slug": "mfpro-nav-batch-calls-and-guides-aug2026",
"title": "Running Hundreds of API Calls to Pull NAV History, One Scheme Per Call? Batch 50 Into One.",
"tags": [
"mutual-funds",
"database-ai"
],
"content": "Running Hundreds of API Calls to Pull NAV History, One Scheme Per Call? Batch 50 Into One.\n\nRunning hundreds of API calls to pull NAV history, one scheme per call?\n\nYou can batch 50 schemes into a single call, which turns two hundred funds into four calls, and if you want everything there is a single download of the full database, April 2006 onward. Same data, far less waiting.\n\nGuide here: Pulling more, calling less\n\nUseful guides and links\n\nWhy am I getting an error, why is there no SIF NAV, why does the AUM look wrong. tigzig.com/post/mf-nav-api-faq-aug2026\n\nA plain guide to what each error actually means, across all three APIs. tigzig.com/post/tigzig-api-errors-practical-guide-jul2026\n\nWhat the fields mean and how it is pulled. The data dictionary, every field with its caveats: tigzig.com/mfpro ➜ Docs\n\nEvery scheme and its latest NAV, one small file. tigzig.com/post/mfpro-scheme-master-download-aug2026\n\nA checker that tells you whether the ISIN is wrong or the data is simply absent. isin.tigzig.com\n\nSIF data. sif.tigzig.com\n\nThe tool itself is at tigzig.com/mfpro, and the API docs at tigzig.com/apis/mf-nav\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, MF File Converter, DUCKIT - CSV to DuckDB\n\nExplore: MF NAV API docs, API and MCP catalog, Python in Excel with xlwings Lite\n\nMore posts: Pulling Indian Mutual Fund NAVs One Scheme at a Time? Three Ways to Get the Same Data in Far Fewer Calls., Three Upgrades on the India Mutual Fund NAV API. Easier to Pull, Whatever You Already Use., MF NAV API FAQ. Why Am I Getting an Error, Why Is There No SIF NAV, and Why Does the AUM Look Wrong?, MFPRO Scheme Master: Every India Mutual Fund Scheme in One Small File, Free, An ISIN Checker for When a Mutual Fund Lookup Fails. And Why One Fund Carries Ten Identifiers."
},
{
"slug": "mfpro-nav-fewer-calls-batch-files-parquet-aug2026",
"title": "Pulling Indian Mutual Fund NAVs One Scheme at a Time? Three Ways to Get the Same Data in Far Fewer Calls.",
"tags": [
"mutual-funds",
"duckdb"
],
"content": "Pulling Indian Mutual Fund NAVs One Scheme at a Time? Three Ways to Get the Same Data in Far Fewer Calls.\n\nIf you are pulling Indian mutual fund NAVs from the MFPro API one scheme at a time, there are three ways to get the same data in far fewer calls, and all three are already in the API.\nYou can batch up to 50 schemes in a single call, and you can pass a date range along with it, so a handful of calls will usually cover what you would otherwise do in a few thousand.\nIf you want the current position across the whole universe, there is one file with the latest NAV for every scheme, around 38,000 of them. A good share of those are closed or matured schemes, and I have left them in rather than deciding on your behalf what counts as live, so you can filter down to the active ones yourself.\nAnd if you want the full daily history for everything, that is one file as well, a little over 37 million records going back to April 2006.\n\nAll updated 3x a day.\n\nDocs are at tigzig.com/apis/mf-nav.\n\nWhat the docs cover, if you are deciding between the three\n\nIf you are deciding between the three, the docs go into the detail. The data dictionary lists every field in both files, how the data is built covers the sources and the three-times-a-day sync, and there is now a guides and FAQ page that collects the questions that come up most, including why a lookup returns nothing and why an AUM figure can look lower than the number published elsewhere.\n\nAll of it sits under Docs on the MFPRO App: tigzig.com -> MFPRO\n\nAnd every page has a plain markdown version if you would rather hand it to an AI agent than read it yourself.\n\nA fourth way, if you do not want the file on disk at all\n\nAnd there is a fourth way if you do not want the file on disk at all. The full history file is parquet, and parquet supports range requests, so DuckDB or Polars can query it straight off the URL and pull only the bytes they actually need. Counting all 37 million rows moved under half a megabyte, and pulling one fund's complete history moved about 1.3 megabytes, out of a file of roughly 175 megabytes.\n\nDetailed writeup here: Range requests on the MFPRO NAV file\n\nPutting some numbers on it\n\nTo put some numbers on it, if you want every active fund's latest NAV each morning, that is about 8,600 schemes. One at a time is 8,600 calls, batched at 50 it comes down to about 172, and the latest file is a single request, roughly 10 MB as CSV or about 1 MB if you take the gzipped version. The data is the same in all three cases.\n\nWhat counts as active, and how to check whether anything changed\n\nOn active, I work it out from the data, and a scheme counts as active if it has reported a NAV in the last 45 days. I kept that window much wider than it needs to be as a buffer and you can apply a further cut based on the nav date.\n\nAnd if you are checking each day whether anything has changed, the files and the manifest both support conditional requests, so if you send If-None-Match or If-Modified-Since with what you already have you get a 304 back with no body at all, and you only pull the file on the days there is something new in it.\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, MF File Converter, DUCKIT - CSV to DuckDB\n\nExplore: MF NAV API docs, API and MCP catalog, Claude for Analytics hub\n\nMore posts: Range Requests on the MFPRO NAV File Are Fixed. DuckDB and Polars Queries No Longer Get Refused Midway., MF NAV API FAQ. Why Am I Getting an Error, Why Is There No SIF NAV, and Why Does the AUM Look Wrong?, MFPRO Mutual Fund API: Search Now Filters Across 12 Fields on a Scheme, Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users"
},
{
"slug": "mfpro-nav-into-excel-aug2026",
"title": "Pull India Mutual Fund NAVs Straight Into Excel. 50, 100 or More Schemes, in a Few Clicks.",
"tags": [
"mutual-funds",
"excel"
],
"content": "Pull India Mutual Fund NAVs Straight Into Excel. 50, 100 or More Schemes, in a Few Clicks.\n\nWant to pull India Mutual Fund NAV's directly into Excel. Say for 50 or 100 schemes. .. or more? Now do that in just a few clicks on MFPRO .. in batches of 20 ..simple interface, advanced search and filters. Free. No Sign In. All schemes from 2006, updated 3x a day.\n\nGet a CSV. Open it in Excel and it is ordinary rows.\n\nRatios. Returns, CAGR, drawdown, whatever you normally work out with formulas.\n\nAI assistance. If you use Claude or something similar alongside Excel, it will write those formulas for you off the same file. If you are using Claude Cowork or ChatGPT Work it will fetch the data for you and do the calculations for you too.\n\nLink: tigzig.com ➜ MFPRO ➜ Data & API\n\nDocs, Guides and Data Dictionary. What each field means, how the data is pulled and how it is checked. Every doc page has a copy button, so you can hand the whole thing to your AI.\n\nFull database. And if you want everything rather than a selection, the full history is one download, April 2006 onward. That one is 37 million rows, so it is not an Excel file, and you would bring it in through something like xlwings Lite reading it via DuckDB.\n\nSome quick points for analysts\n\nName changes. And they have changed a lot this year. SEBI reissued the categorisation framework in February and AMCs have been re-filing since, so the same fund can turn up under a new category name, and sometimes a new scheme name, between two of your files. If you are matching on name, that break shows up as rows dropping out of your join without any error.\n\nUse the AMFI scheme code as your key. It survives renames. ISIN works too, but check your coverage before you rely on it: about one in fourteen live schemes carries no ISIN at all, and across the full history including matured schemes it is closer to a quarter. Some schemes also carry a second ISIN for the reinvestment option.\n\nDirect and Regular are separate scheme codes, and so are Growth and IDCW, so one fund in your head is usually four rows in the file.\n\nSEBI has revised mutual fund categorisation\n\nSIF is separate for now\n\nSIF data is not yet integrated into MFPRO. It sits separately on sif.tigzig.com. Small file .. ~11K rows .. just download the whole thing into Excel.\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, DUCKIT - CSV to DuckDB, MF File Converter\n\nExplore: MF NAV API docs, Python in Excel with xlwings Lite, Claude for Analytics hub\n\nMore posts: SEBI Has Revised Mutual Fund Categorisation and Introduced a New Fund Type. The Major Changes, and What They Mean for MFPRO Users and Analysts., MF NAV API FAQ. Why Am I Getting an Error, Why Is There No SIF NAV, and Why Does the AUM Look Wrong?, Pulling Indian Mutual Fund NAVs One Scheme at a Time? Three Ways to Get the Same Data in Far Fewer Calls., An Excel Tracker for Hundreds of Yahoo Finance Tickers, Built with xlwings Lite. For When WEBSERVICE Keeps Failing."
},
{
"slug": "mfpro-range-requests-duckdb-polars-aug2026",
"title": "Range Requests on the MFPRO NAV File Are Fixed. DuckDB and Polars Queries No Longer Get Refused Midway.",
"tags": [
"mutual-funds",
"duckdb",
"infrastructure"
],
"content": "Range Requests on the MFPRO NAV File Are Fixed. DuckDB and Polars Queries No Longer Get Refused Midway.\n\nA number of you have been running duckdb or polars queries against the tigzig MFPRO NAV file over the past couple of weeks and getting blocked midway through, and that is now fixed.\n\nThe reason turned out to be simple. One analytical query fires off close to three hundred separate byte range requests in a few seconds, and the rate limiting in front of the API was counting those range reads in the same bucket as ordinary whole file requests, so a legitimate query looked like a burst and got refused halfway through.\n\nRange requests now have their own pathway and their own budget, so you are free to run queries against my server without that happening again. The pathway supports up to 3,000 requests per minute and 25,000 per day per (public) IP. And response headers come with remaining limit details.\n\nThe file itself is about 175 MB with close to 37 million NAV records, and it is still there to download whole if you would rather.\n\nThe API also continues to support single and batch NAV pulls, search across scheme fields, and the newer snapshot endpoint, which gives the latest NAV for every scheme in one file instead of the full history just for today's number.\n\nLinks:\nMFPRO App: tigzig.com/mfpro\nMF Data API, for humans: api.tigzig.com/mf/v1/docs\nMF Data API, for agents: api.tigzig.com/mf/v1/openapi.json\nFull catalog (agents): api.tigzig.com\n\nTips\n\nLibrary comparison\n\nduckdb and polars are the fastest. Polars slightly faster than duckdb in our tests. pyarrow and pandas (with fsspec) work fine but move roughly 10x more than duckdb/polars for the same query. pandas given a bare URL is the one to avoid outright.\n\nThe duckdb snippet\n\n``sql\nINSTALL httpfs; LOAD httpfs;\nSELECT date, nav FROM read_parquet('https://api.tigzig.com/mf/v1/download?format=parquet')\nWHERE scheme_code = 122639 ORDER BY date DESC;\n`\n\nAbout 3,250 rows, full history for one fund since 2013, about 1.3 MB moved, a few seconds.\n\nThe filter column\n\nThe file is sorted by schemecode. WHERE on schemecode costs about 5 requests. A WHERE on date, or no filter at all, costs about 300, because every row group has to be opened.\n\nThe pandas gotcha\n\nDon't pass a bare url.\n\n`python\nslow: 372 MB moved, 37 seconds, for 3,245 rows\ndf = pd.read_parquet(URL, columns=[...], filters=[...])\n\nfast: 17 MB, 3.5 seconds, same rows\nimport fsspec\nh = fsspec.filesystem(\"http\").open(URL)\ndf = pd.read_parquet(h, columns=[...], filters=[...])\n``\n\nQuery vs download\n\nFull file download: about 36 seconds. A schemecode-filtered query: a few seconds. An unfiltered scan or a name search: worse than just downloading the whole file. It all depends on the specific use case and your constraints .. rough rule, filter on schemecode and query remotely, anything else, just download it or use the regular API... unless you have a specific reason to do a range request. Based on the volume and variety of requests that I see on the backend, I might optimize the parquet further.\n\nJavaScript / Node options\n\nDoesn't have to be Python. Sharing below two options which I tested lightly only ..might be more ...\nhyparquet: pure JS, zero deps, works with plain fetch. On a simple read it was fine. On a filtered query against the big file it did not prune the way duckdb does, it fired several hundred concurrent range requests at once and one connection timed out. Usable, but rougher than the Python/duckdb path, needs care.\nnative \"duckdb\" npm package (not duckdb-wasm): same engine as Python's duckdb, worked cleanly first try, same smart pruning. Best option if your JS runs somewhere that allows native code (a normal Node server, Lambda).\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: MF NAV API docs, Mutual fund analytics hub, API + MCP hub\n\nMore posts: MFPRO Scheme Master: Every India Mutual Fund Scheme in One Small File, Free, Claude Cowork Runs a Real Python Sandbox. I Put 37 Million Rows of Mutual Fund NAV Data Through It., 1 in 5 Bulk Downloads Were Silently Truncating. Two Cloudflare Worker Bugs, Found and Fixed. Now Up to 4x Faster., Zero Callers Rate Limited in 24 Hours. 6,500+ Calls and 9.7 GB Served, After the TigZig API Limits Went Up., 227 Downloads, 41GB in 10 Days. India MF Complete NAV Database - 38K+ Schemes, 37M+ Records, One 167MB Parquet. Free."
},
{
"slug": "mfpro-scheme-details-live-oct2026",
"title": "Scheme Details for Indian Mutual Funds Are Live on Tigzig: Fund Size, Fund Manager, TER and BER, Benchmark and Riskometer",
"tags": [
"mutual-funds",
"database-ai"
],
"content": "Scheme Details for Indian Mutual Funds Are Live on Tigzig: Fund Size, Fund Manager, TER and BER, Benchmark and Riskometer\n\nScheme details for Indian mutual funds are now live on Tigzig APIs, tool and data surfaces - available for people and agents - fund size, fund manager, TER and BER, benchmark, riskometer and scheme terms. Refreshed 4 times a day.\n60 fields per scheme, in one file and one API.\nCovering ~2100 funds across 8600+ plans.\nPull one scheme, up to 50 in a call.\nDownload the whole file as Parquet or CSV.\nData quality file: mismatches between sources, with the reason. Available on the API.\nValidations: on the MFPRO Data Quality page, covering NAV, scheme documents and holdings.\n\nIt adds to the full NAV history already on MFPRO: about 37.5 million daily NAVs for 38,000+ schemes, live and matured, since April 2006. NAVs work the same way, one fund at a time, 50 in a batch, the latest NAV only, or the whole database as one download.\n\nFree, no signup, no API key.\n\nIt's actually been live for the past few days and it is already being pulled into Python scripts, Google Sheets and people's own apps.\n\nFor you: https://www.tigzig.com/mfpro\n\nFor your AI: https://api.tigzig.com/mf/v1/openapi.json\n\nUsage notes\n\nThere are many data nuances. I have captured them in API docs and on the Data Quality page in MFPRO, with its validations and known issues. A few usage tips below.\nFund-level values repeat on every plan. AUM, fund manager, benchmark and the scheme terms belong to the fund, so they show up on each of its plans, four rows for a typical fund and up to 36 for the largest. To total AUM, count only the rows where funddedupflag is true.\nJoin on schemecode. The file has one row per schemecode, the same code the NAV API and downloads use, so you join on scheme_code.\nAUM latest and AAUM are different numbers. AUM latest is the whole fund on one day, from the daily fund performance data on AMFI's website. AAUM is the quarterly average for each plan, from AMFI's quarterly release, and it stays in the scheme master file as before.\nThere are three riskometers: the daily one for the scheme, the same for its benchmark, and the one the fund house prints in its scheme document. They usually agree, and all three are kept as filed.\nTER is matched only on the exact SEBI scheme code in the fund's own document. Where that code is missing or is shared with another fund, TER is left blank.\n\nhttps://www.tigzig.com/apis/mf-nav\n\nUseful links for Tigzig MF API users and agents\nSame data in fewer calls.\nhttps://www.tigzig.com/post/mfpro-nav-fewer-calls-batch-files-parquet-aug2026\nDemystifying TIGZIG API Errors.\nhttps://www.tigzig.com/post/tigzig-api-errors-practical-guide-jul2026\nMF NAV API FAQ.\nhttps://www.tigzig.com/post/mf-nav-api-faq-aug2026\nTigzig API Status page.\nhttps://status.tigzig.com\nTigzig ISIN Checker.\nhttps://isin.tigzig.com\nTips: https://www.tigzig.com/post/isin-checker-mutual-fund-identifiers-aug2026\n\nScheme details for Indian mutual funds, live on Tigzig\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, QRep - Security Reports, QDesk - Quant Report Desk\n\nExplore: Mutual funds hub, MF NAV API docs, What is new\n\nMore posts: Pulling Indian Mutual Fund NAVs One Scheme at a Time? Three Ways to Get the Same Data in Far Fewer Calls., MF NAV API FAQ. Why Am I Getting an Error, Why Is There No SIF NAV, and Why Does the AUM Look Wrong?, An ISIN Checker for When a Mutual Fund Lookup Fails. And Why One Fund Carries Ten Identifiers., Heavy User of the Mutual Fund NAV API? Run Your Own NAV Service Inside Your Own Setup, Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users"
},
{
"slug": "mfpro-scheme-explorer-sebi-categories-jul2026",
"title": "MFPRO Scheme Explorer is Live. Cross-Tab 38,000+ India Mutual Fund Schemes, and Watch the SEBI Re-Categorisation Happen in the Daily Files.",
"tags": [
"mutual-funds"
],
"content": "MFPRO Scheme Explorer is Live. Cross-Tab 38,000+ India Mutual Fund Schemes, and Watch the SEBI Re-Categorisation Happen in the Daily Files.\n\nWant to see the actual distribution of Indian mutual funds? With new mandated SEBI categories and names. How many active? by category / sub-category? By AMC? stale, closed, active etc etc..\n\nI have set up an interactive Scheme Explorer on MFPRO where you can cross-tab the entire universe of 38,000+ schemes. You can slice and pivot scheme counts by clean categories, subcategories, AMC, plans, and options. And also apply filters as you want ....\n\nFor the SEBI category and name changes (Feb '26 Circular, last date to implement for AMC's is late Aug'26) ....read more here: SEBI has revised mutual fund categorisation and introduced a new fund type\n\nWhat the re-categorisation looks like in the daily files\n\nAnd it is already flowing through AMFI's daily NAV files, and since MFPRO now tracks the latest filing, you can watch it live.\n367 schemes changed their category or name in a single day, including 91 scheme name changes as fund houses rename to match the new true-to-label rule.\nThe renames are exactly what the circular asked for: Groww Dynamic Bond Fund is now Groww Dynamic Term Fund, ITI Ultra Short Duration Fund is now ITI Ultra Short Term Fund, ICICI Prudential Equity Arbitrage Fund is now just ICICI Prudential Arbitrage Fund.\nSome renames go beyond the mechanical: Canara Robeco Consumer Trends Fund and HDFC Non-Cyclical Consumer Fund are both now simply Consumption Fund.\nThe first two Life Cycle Funds are live, Zerodha Life Cycle Fund 2036 and 2041, the new target-maturity category that starts equity heavy and shifts to debt as the year approaches.\n\nTry pivoting the distribution yourself\n\ntigzig.com/mfpro -> Explorer\nWant to look up a Fund? Then go to the 'Fund' tab.\nCompare up to 200 funds on risk, return, benchmark metric -> that's the home page\nHow is it calculated? Definitions? Go to Docs\nWant to analyze yourself offline? Download the whole 37M+ rows of data -> go to Data & API\n\nSome changes that caught my eye\nICICI Prudential Thematic Advantage Fund (FOF) is now ICICI Prudential Aggressive Hybrid Active FOF\nThe new FoF naming convention is visible too: Quantum Gold Savings Fund is now Quantum Gold ETF FOF, and Motilal Oswal Asset Allocation FOF is now Motilal Oswal Multi Asset Passive Fund of Fund. The name now tells you whether the underlying funds are active or passive, which the old names never did.\nChildren's funds and retirement funds are on their way out as a category. SEBI discontinued Solution Oriented schemes, so the 143 active ones stop taking fresh subscriptions and will be merged into similar schemes. If you hold one, watch for the merger notice from your fund house.\nUTI Liquid Cash Plan, a name from another era, is now simply UTI Liquid Fund.\n\nOf the 38,000+ schemes in the database, only about 8,600 are active. Roughly three out of four schemes that ever existed have matured or merged, most of them fixed maturity plans from the 2006 to 2013 era.\n\nOne quirk I spotted\n\nAditya Birla Sun Life Gold ETF is currently showing under \"Equity ETF\" in AMFI's daily file. The fund itself has not changed, it still holds gold. This looks like a transient filing error during the re-categorization rush and will most likely be corrected in a subsequent filing. We publish the data as filed, so you can see it in the Explorer today, and you will see it flip back when it gets corrected.\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Mutual fund analytics hub, MF NAV API docs, API + MCP hub\n\nMore posts: SEBI Has Revised Mutual Fund Categorisation and Introduced a New Fund Type. The Major Changes, and What They Mean for MFPRO Users and Analysts., Data Dictionary & Docs Live on MFPRO. Every Field Explained, 21-Metric Methodology, and How the India MF Universe Actually Looks., One Line of User Feedback, Fixed Within a Day. MFPRO Search Upgraded + Complete AMFI NAV History, 37M Rows, 2006 Onward., MFPRO Mutual Fund API Update. Three Things Your Calls Asked For, Now Built."
},
{
"slug": "mfpro-scheme-master-download-aug2026",
"title": "MFPRO Scheme Master: Every India Mutual Fund Scheme in One Small File, Free",
"tags": [
"mutual-funds",
"portfolio-analytics"
],
"content": "MFPRO Scheme Master: Every India Mutual Fund Scheme in One Small File, Free\n\nOne row for every mutual fund scheme in AMFI's public records from April 2006 onward, all 38,000 of them, with the fund's name, AMC, category, plan, option, first and last NAV dates, its latest average AUM and its latest NAV. About 9 MB as plain CSV, 1 MB compressed, no key and no sign-up. Take it as a one-click download from the app, or fetch it from code, whichever suits you. It opens in Excel. Paired with the full NAV history download, it gives you the same universe that runs the MFPRO app itself.\n\nWhat this is\n\nMFPRO's NAV API has always been able to give you a fund's full price history once you know its scheme code. The gap was everything around the scheme, what it is called today, who runs it, which category it sits in, whether it is still alive, how big it is. That information was reachable through search, one query at a time.\n\nThe scheme master is all of it in one file, and you have the whole scheme universe on your machine to slice however you want.\n\nHow do I get it?\n\nTwo ways, and they hand you exactly the same file. Pick whichever suits how you work.\nClick and download it\n\nNo code needed at all. Go to the MFPRO app, Data and API tab, and under Download Scheme Master pick your format: CSV at about 9 MB, zipped CSV at 1.1 MB, or Parquet at 1.6 MB. One click, and it saves like any other file. The plain CSV opens straight in Excel, so if all you want is to sort and filter 38,000 schemes in a spreadsheet, you are done at that point. The full NAV history sits on the same page under Download Full Database, in CSV, TSV, Parquet or SQLite.\nCall it from code\n\nSame files, fetched over HTTP, no key and no sign-up:\n\n``\nhttps://api.tigzig.com/mf/v1/download?format=latest\n`\n\nThat is the plain CSV. Add latest.csv.gz, latest.csv.zip or latest.parquet for the other three. pd.readcsv and pd.readparquet both work directly off the URL with no download step, which makes it a one-line refresh inside a script or a notebook.\n\nEverything below this point applies whichever route you took. The columns, the quirks and the tips are about the data itself rather than about how you fetched it.\n\nWhat you get in it\n\n22 columns per scheme, and they group into five ideas:\nWho it is. schemecode (AMFI's number, the key you use everywhere else in this API), isin and isin2, where isin2 is the IDCW reinvestment ISIN for the schemes that have one, the same scheme and the same NAV series, so a lookup works with either. Then schemename as filed today, amc, and txic_code, our own readable short code.\nWhat it is. schemetype (open, close, interval), the full raw category text as AMFI files it, categorysub (the SEBI sub-category, Small Cap Fund and so on), categorygroup (AMFI's own group header) and categorygroup_clean, which is the tidy four-value version, Equity, Debt, Hybrid, Other.\nHow it is sold. schemeplan (Direct, Regular, Other) and schemeoption (Growth, IDCW, Bonus, Other).\nIts lifespan. firstdate and lastdate, the first and last dates it published a NAV, plus isactive and isstale.\nIts size and price. aaumcrquarterlyavg, the latest quarterly average AUM in crore, with aaumquarter and aaumquarterend telling you which quarter, and nav_date plus nav for the latest published NAV.\n\nAbout 8,600 of the schemes are live. The other 29,000 or so have matured or merged, and they are in the file on purpose, carrying their final NAV frozen at the day they stopped publishing. That is what lets you run a long-run study on this data survivorship-bias free, with the funds that died still in the sample.\n\nHow you would actually use it\n\nCan I do all of this in Excel, without writing any code?\n\nFor most of it, yes. Download the CSV, double-click, and you have 38,000 rows and 22 columns you can filter and sort like any other sheet. Filter isactive to TRUE, categorygroupclean to Equity Scheme, schemeplan to Direct, sort by aaumcrquarterlyavg descending, and you have the biggest active Direct equity schemes in the country in about thirty seconds, with their scheme codes in the first column ready to paste wherever you need them. Questions like \"which AMCs run Small Cap funds\" or \"how many Direct Growth schemes exist\" are a filter and a row count. The one thing worth doing in a formula rather than a filter is track-record length, since that is navdate minus first_date.\n\nI want a subset before I do any real analysis. How?\n\nThis is what the file is best at. Say you want every active equity scheme from three fund houses, above a size threshold, with at least ten years of published history, Direct plan, Growth option. That is one filter step:\n\n`python\nimport pandas as pd\ndf = pd.read_csv(\"https://api.tigzig.com/mf/v1/download?format=latest\")\n\ndf[\"firstdate\"] = pd.todatetime(df.first_date)\ndf[\"navdate\"] = pd.todatetime(df.nav_date)\ndf[\"years\"] = (df.navdate - df.firstdate).dt.days / 365.25\n\npicks = df[(df.is_active) &\n (df.categorygroupclean == \"Equity Scheme\") &\n (df.amc.str.contains(\"HDFC|ICICI|SBI\", case=False, na=False)) &\n (df.aaumcrquarterly_avg > 500) &\n (df.years >= 10) &\n (df.scheme_plan == \"Direct\") &\n (df.scheme_option == \"Growth\")]\n\ncodes = picks.scheme_code.tolist()\n`\n\nThat returns 39 schemes carrying about Rs 2.9 lakh crore between them, HDFC Flexi Cap, HDFC Mid Cap, ICICI Prudential Large Cap, SBI Contra and the rest. Now you have your scheme codes, and you go get the NAVs.\n\nOne note on \"ten years of history\"\n\nThe file carries first_date, the first date a scheme published a NAV in AMFI's records, which is not quite the same thing as the fund's launch date. For a fund launched after April 2006 the two are close enough to use as a proxy for track-record length, which is how the filter above uses it. For a fund that predates that, the NAV record starts in April 2006 where AMFI's published archive starts, so read the column as \"how much history you can actually analyse\" rather than \"how old the fund is\". For a backtest, the first of those is the number you want anyway.\n\nI have my scheme codes. What next?\n\nThree routes, depending on how many you picked.\nA handful. Call /nav one scheme at a time, or add latest=true if you only want today's number rather than the whole series.\nA basket. Up to 50 codes in a single call with ?schemes=code1,code2,..., which is what you want for a portfolio.\nA few hundred, or the lot. Stop calling the API and take the full history file instead, ?format=parquet, about 170 MB, 37 million rows, every scheme, every day back to April 2006.\n\nCan I skip the API completely?\n\nYes. Both files are static downloads. Take them once, load them into DuckDB, SQLite, pandas, Excel, whatever you use, and query offline as much as you like. The files are rebuilt three times a day, so re-download when you want fresh numbers.\n\nScheme Master + Full NAV History = the whole universe\n\nThe two downloads are built to be used together, and this is the combination worth setting up once.\n\nThe scheme master is the index. 38,000 rows, about 9 MB, one row per scheme, everything you need to decide which funds you care about. The full NAV history is every price ever published, 37 million rows back to April 2006, about 170 MB as parquet.\n\n`python\nimport pandas as pd\n\nmaster = pd.read_csv(\"https://api.tigzig.com/mf/v1/download?format=latest\")\nnavs = pd.read_parquet(\"https://api.tigzig.com/mf/v1/download?format=parquet\")\n\ndf = navs.merge(master, on=\"scheme_code\")\n`\n\nJoin them on scheme_code and you have, sitting locally, the same universe the MFPRO app itself runs on. No rate limit, no pagination, no network in the loop, and no scraping AMFI. Rolling returns, drawdowns, category comparisons, AMC-level rollups, survivorship-free backtests that include the funds that died, all of it becomes a query against your own machine.\n\nThe reason to take both rather than only the big one is that the scheme master is what makes the big file usable. Filtering 38,000 rows to pick your scheme codes takes no time, whereas searching 37 million NAV rows for the same funds by name, with no category or AMC column to filter on, is slow and easy to get wrong.\n\nQuirks worth knowing before you trust a number\n\nEvery one of these produces a plausible-looking wrong answer rather than an error, which is why they are worth reading once.\n\nThe NAV column is not \"today's NAV\" for every row\n\nIt is each scheme's most recent NAV. For the 29,000 matured schemes, that is their final NAV from whenever they wound up, sometimes ten or fifteen years ago. Take an average NAV across the file and you have mixed 2026 prices with 2011 prices. Filter isactive, and read navdate to see how fresh each row is.\n\nOne scheme code is one plan-and-option variant, not a fund\n\nThis one catches everybody once. Aditya Birla Sun Life Value Fund is four rows in this file, two Growth and two IDCW, spread across Direct, Regular and an older pre-2013 code that classifies as plan Other, each row with its own scheme code and its own slice of the AUM. Quote any single row as \"the fund's size\" and you understate it badly, in that case by about nine tenths. To size a fund, group by AMC and base name and sum. To count funds rather than variants, do the same before counting.\n\nBlanks are common, and they usually mean something\n\nA blank is rarely missing data here. categorysub is blank for a large share of the file, roughly 25,000 rows, because AMFI never assigned a modern SEBI sub-category to old-style schemes from before 2018, the ones filed simply as Income or Growth. That is by design, and their raw label is still sitting in the category column. Similarly aaumcrquarterlyavg is blank for almost all matured schemes, since AMFI stops publishing an AUM once a fund is gone, and isin is blank for about 9,000 mostly old schemes that never had one recorded. If a filter on categorysub returns far less than you expected, blanks are usually why, so filter on categorygroup_clean instead when you need to include the old schemes.\n\nSort quarters on the ISO column, never the label\n\naaumquarter is a text label like \"June-2026\", and text sorts alphabetically, which puts December before June before March. aaumquarter_end is the same quarter as a proper date, 2026-06-30, and sorts correctly. Sort and filter on the date, display the label. Also worth knowing that schemes legitimately sit on different quarters at any one time, because AMFI publishes each new quarter fund house by fund house over several weeks.\n\nAAUM is an average, not a balance\n\naaumcrquarterly_avg is the SEBI-defined average of daily net assets across the quarter, so it will not match the \"current AUM\" on a factsheet, and a fund launched mid-quarter reads low against a full-quarter peer because it is averaged over all the quarter's days regardless.\n\nA small number of matured schemes carry a final NAV of zero\n\nAbout 2,300 rows, all of them matured, show a last NAV of exactly 0.00, roughly 8 percent of the matured universe. That zero is AMFI's own closing entry for the scheme, so read it as the end of the series rather than as a price. The daily sync filters zero values out of the live feeds, and the historical backfill kept archive values as AMFI filed them, which is why these closing rows are still here. A straight \"latest NAV\" pull across matured funds will pick up those zeroes, so screen them out before computing anything from that column.\n\nRecent changes elsewhere in the MFPRO API\n\nThe scheme master is one of several things that moved this month. Short version of the rest, with links if any of it is useful to you.\n\nSearch now filters on 12 fields of its own\n\nIf you would rather query than download, /search now filters on AMC, category, group, type, plan, option, active, stale, launch and last-NAV date windows, ISIN and TXIC code, with commas as OR, filters ANDing together, and an exclude that works as NOT. The write-up covers how they combine, what a wrong call returns, and the real limits. Search now filters across 12 fields on a scheme\n\nThree things the logs asked for, now built\n\nlatest=true is a real parameter now rather than something callers kept guessing at, identifier intake got a lot more forgiving, and responses carry freshness metadata so a poller can be cheap. Three things your calls asked for, now built\n\nGetting an error? Start here\n\nA practical guide to what actually goes wrong when people call these APIs, built from real server logs rather than theory. Trailing full stops copied into URLs, encoded ampersands, wrong parameter names borrowed from another API, browsers used where a POST was needed, and the difference between a failed call and an empty result. It covers MFPRO, TREMOR, VIGIL and the Yahoo Finance tools together, since the same handful of mistakes repeat across all of them. Demystifying TigZig API errors\n\nWhat other people are doing with these APIs\n\nThe server logs rewritten as a plain-English conversation: who calls, what they ask for, the patterns worth copying and the ones worth avoiding. Useful mostly as a shortcut to good practice. Your agents and scripts have been talking to me\n\nThe bulk downloads got faster, and a real bug got fixed\n\nRoughly one in five bulk downloads had been truncating, files that started fine and arrived incomplete with nothing shown to the user. Two Cloudflare Worker bugs, found and fixed, and the downloads are now up to four times quicker. 1 in 5 bulk downloads were truncating\n\nSEBI has rewritten fund categories and names\n\nRelevant to anything you build on the category columns: SEBI's February 2026 circular renamed several categories, split Sectoral from Thematic, added Life Cycle Funds and requires fund names to match their category by 26 August. Old and new category names coexist in the data during the transition. What SEBI changed and watching it happen in the daily files\n\nEvery field, explained properly\n\nThe full data dictionary covers every column with its caveats, alongside the scheme universe field guide and a page on how the data is assembled. Data dictionary, scheme universe field guide, how the data is built\n\nEverything here is open, no key and no sign-up. The MF NAV API guide has the full endpoint reference, the MFPRO app has a Scheme Explorer if you would rather pivot the universe in a browser than in code, and there is a Report button in the app header if something here does not match what you find.\n\nAI agents: the machine-readable version of everything above is at https://api.tigzig.com/mf/v1/, with per-column semantics under schema.latestsnapshot.columnnotes and the OpenAPI spec at https://api.tigzig.com/mf/v1/openapi.json`.\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: MF NAV API docs, Mutual fund analytics hub, API + MCP hub\n\nMore posts: MFPRO Mutual Fund API: Search Now Filters Across 12 Fields on a Scheme, 227 Downloads, 41GB in 10 Days. India MF Complete NAV Database - 38K+ Schemes, 37M+ Records, One 167MB Parquet. Free., MFPRO Scheme Explorer is Live. Cross-Tab 38,000+ India Mutual Fund Schemes, and Watch the SEBI Re-Categorisation Happen in the Daily Files., Data Dictionary & Docs Live on MFPRO. Every Field Explained, 21-Metric Methodology, and How the India MF Universe Actually Looks., 1 in 5 Bulk Downloads Were Silently Truncating. Two Cloudflare Worker Bugs, Found and Fixed. Now Up to 4x Faster."
},
{
"slug": "mfpro-search-filters-12-fields-jul2026",
"title": "MFPRO Mutual Fund API: Search Now Filters Across 12 Fields on a Scheme",
"tags": [
"mutual-funds",
"infrastructure"
],
"content": "MFPRO Mutual Fund API: Search Now Filters Across 12 Fields on a Scheme\n\nMFPRO is TigZig's India mutual fund NAV and analytics API. Its search used to match a fund by name or its scheme code and nothing past that. Now it filters on 12 fields on a scheme, six text fields, two booleans, two date windows and two identifiers, plus exclude sitting on top of all of them as the NOT. Every one of the 12 combines with the others. This piece covers the full list, how the filters actually combine, and what a wrong call gets back.\n\nWhat this endpoint is for\n\nThe rest of this API works off scheme codes. /nav and /schemes/{code}/nav return a fund's NAV history, but only once you already have the code, or the ISIN, to ask for. Most people do not start there. They start with a fund name, or with a rough idea of what they want, every Direct Growth Small Cap fund from one AMC, say, not a code at all.\n\n/search is how you get from one to the other. Give it a name and it finds the matching schemes. Give it filters instead, category, AMC, plan, launch date, whatever you know, and it shortlists the schemes that match, codes included. Most pipelines built on this API run in two steps because of this: search first to get the codes, then call /nav or /schemes/{code}/nav on whatever came back.\n\nIt is also useful without a second step at all, just to see the shape of the universe, how many schemes sit in a category, how many an AMC runs, how many are Direct versus Regular. That kind of browsing, and everything else covered below, needs no NAV call after it.\n\nWhat it used to do\n\nFor a long stretch the endpoint took one parameter that did anything, q, a fund name or a scheme code, matched as a substring, case insensitive. Ask for hdfc flexi cap and you got every scheme with those words somewhere in its name, ranked by how close the match was. That part still works exactly the same way, and it is still the fastest route in when you already know roughly what fund you want.\n\n``\nGET https://api.tigzig.com/mf/v1/search?q=hdfc flexi cap\n`\n\nFour schemes come back, the growth and IDCW options across HDFC Flexi Cap Fund's direct and regular plans, each carrying a relevance score, with the direct growth plan on top at 26 against 21 for the other three.\n\nWhat it does now\n\nThe same endpoint now reads 12 fields on a scheme, and you can filter on any of them with no name at all. Give it filters and nothing else and it browses the scheme universe for you, the way you would browse a catalog rather than search one, active=all&limit=5 with no q at all returns the first five schemes out of more than 38,000. There is no separate switch for this and no q=all, any filter present is what turns search into browse. q stays mandatory only on a completely bare call, no filters and no q, which still returns a 422. Here is the full list.\n\n| Field | What it matches | Example |\n| --- | --- | --- |\n| amc | AMC name, contains | amc=aditya birla |\n| category | the SEBI sub-category, contains | category=small cap |\n| group | the four broad buckets: Equity Scheme, Debt Scheme, Hybrid Scheme, Other Scheme | group=equity |\n| type | open, close or interval schemes | type=open |\n| plan | Direct, Regular or Other | plan=direct |\n| option | Growth, IDCW, Bonus or Other | option=growth |\n| active | true, false or all. Defaults to true on a pure browse, so matured schemes stay out unless you ask for them | active=all |\n| stale | true or false, the roughly 50 schemes that carry only one or two NAV rows in AMFI's own records. All of them are matured, so pair it with active=all or you get nothing back | stale=true&active=all |\n| firstdatefrom / firstdateto | the launch window, when a scheme first published a NAV | firstdatefrom=2015-01-01 |\n| lastdatefrom / lastdateto | the window for when a scheme stopped publishing | lastdateto=2015-12-31 |\n| isin | an exact ISIN, not a substring match, comma separated for more than one | isin=INF174K01LT0 |\n| txic | TigZig's own scheme code, contains | txic=AXOMC |\n\nexclude sits on top of all 12 as the one NOT. It is not a thirteenth field to filter on, it is the operator that removes matches from whatever the other filters already found, and the section below covers it properly.\n\nA note on group, because it moved recently. SEBI's February 2026 circular discontinued Solution Oriented schemes, and the clean grouping went from six buckets to four in the process. Solution Oriented and Fund of Funds now sit inside Other Scheme, and their schemes stay fully findable through category and the raw category group. If you have code written against the older six-value list, group=solution oriented and group=fund of funds now return nothing.\n\ncategory and group are not the same field\n\nThe two fields, in short\n\ngroup is the four broad buckets. category is the specific SEBI sub-category sitting underneath one of those buckets, Small Cap Fund, Liquid Fund, Index Funds, and so on, and it is the one that actually answers a specific question. Reach for the wrong one of the two and a perfectly real set of funds does not turn up.\n\nIndex funds, the plain case\n\ncategory=index returns HDFC Nifty 50 Index Fund, ICICI Prudential Nifty 50 Index Fund, SBI Nifty Index Fund, and well over a thousand more, because the filter is a contains-match against the full category text rather than a lookup against the clean sub-category alone. Most of what comes back genuinely carries a category_sub of Index Funds. The rest are schemes like SBI's Crisil IBX SDL Index Fund, whose category reads \"Index Funds - Debt Funds\", a debt-oriented index product that category=index is right to catch even though its own sub-category says Debt Funds, not Index Funds. group=index returns nothing, because Index is not one of the group names. Every index fund in this database sits inside the Other Scheme group, and group alone cannot tell one apart from the other fund types sharing that bucket. The empty response also hands back the valid group names on its own, the same self-correcting habit the wrong-parameter example further down shows off properly.\n\nWhy the filter reads raw text too\n\nEvery scheme carries two versions of its category. There is the raw text AMFI itself prints, a label like \"Equity Scheme - Small Cap Fund\", or for very old schemes a single word like \"Income\". From that raw text we extract a clean field, category_sub, values like Small Cap Fund, Liquid Fund, Index Funds. About 9,000 pre-2018 schemes were never given one of these clean sub-categories by AMFI at all, only the raw text exists for them.\n\nThat is why category= reads both fields rather than the clean one alone. category=income&active=all proves it, tens of thousands of old Income schemes come back, schemes a clean-field-only filter would miss entirely since none of them carry a modern sub-category to match against. The active=all matters here, because those Income schemes are long matured and the browse default of active=true hides almost all of them. The SBI SDL Index Fund case above is the same mechanism from the other side, a compound raw string picking up a match a clean-only filter would have missed too. There is no separate filter today that reads only the clean field, filter broad and check category_sub on the results if you need the exact one.\n\nThe newer example: Life Cycle Fund\n\nSEBI's Life Cycle Fund category was created by the 26 February 2026 circular on scheme categorization. Two schemes exist against it right now, Zerodha Life Cycle Fund 2036 and Zerodha Life Cycle Fund 2041, carrying category_sub of \"Life Cycle Fund with Maturity of 10 Years\" and \"Life Cycle Fund with Maturity of 15 Years\". Both sit inside the Other Scheme group rather than Equity Scheme, so a search built around group=equity on the assumption that a life cycle fund is basically an equity product misses both of them entirely, group=equity&category=life cycle comes back with zero.\n\nThe spelling trap inside the same example\n\ncategory=lifecycle, written as one word, also returns zero, because the field is contains-matched against the literal text and AMFI's own category string is \"Life Cycle\", two words with a space. category=life cycle, spaced the way the source data actually spells it, returns exactly the two Zerodha schemes above. A category filter that matches nothing does not necessarily mean the category is empty, sometimes it means the spelling guessed was close but not exact.\n\nHow the filters combine\n\nTwo different filters AND together on their own, no keyword needed. Ask for a category and a plan and you get schemes matching both.\n\n`\nGET https://api.tigzig.com/mf/v1/search?amc=birla&category=value\n`\n\nFour schemes come back, all four variants of Aditya Birla Sun Life Value Fund, growth and IDCW across direct and regular. Neither filter alone would have narrowed it to that set, the AMC filter alone would have returned every Aditya Birla scheme and the category filter alone every Value Fund from every AMC.\n\nA comma inside one filter is an OR within that field.\n\n`\nGET https://api.tigzig.com/mf/v1/search?amc=birla,hdfc\n`\n\nThis returns 787 schemes total, every scheme from either AMC.\n\nexclude is the one NOT. It drops any result whose scheme name or AMC contains any of the comma separated terms you give it. Give it twice instead of comma separated and it merges the same way every other list filter does, exclude=birla&exclude=jm and exclude=birla,jm both drop category=value from 96 matches to 88.\n\n`\nGET https://api.tigzig.com/mf/v1/search?category=value&exclude=birla\n`\n\ncategory=value alone returns 96 total matches. Add exclude=birla and it drops to 92, exactly the four Aditya Birla Sun Life Value Fund schemes from the AND example above, gone.\n\nThat is the whole language, three rules, and there is nothing past them. There is no way to write the words AND or OR into the query text itself and have them mean anything. Type them into q and they get read as literal words to match against a fund's name.\n\n`\nGET https://api.tigzig.com/mf/v1/search?q=birla AND value\n{\"query\":\"birla AND value\",\"count\":0,\"results\":[]}\n\nGET https://api.tigzig.com/mf/v1/search?q=birla OR value\n{\"query\":\"birla OR value\",\"count\":0,\"results\":[]}\n`\n\nBoth come back empty, and not because the words AND and OR are being read as logic. The match requires every word in the query to appear somewhere in the scheme name, and that includes the words AND and OR themselves, which is why a plain q=birla value finds all four Birla Value Fund schemes while adding AND or OR to that same query knocks it back to zero, no Birla Value fund name happens to contain the letters \"and\" or \"or\". There are no parentheses either, and no way to nest one group of conditions inside another. That is a real limit of the endpoint as it stands, and there is no alternate syntax that gets around it.\n\nWhen a call is wrong\n\nGet a parameter name wrong and the call does not go through with the bad piece dropped and nothing said about it. Ask for exclude_amc instead of exclude, or categry instead of category, and the whole request is rejected.\n\n`\nGET https://api.tigzig.com/mf/v1/search?category=value&exclude_amc=birla\n\n{\"detail\":\"Unknown parameter(s): excludeamc. Search parameters are strict so a typo can never silently return unfiltered results.\",\"error\":{\"code\":\"UNRECOGNIZEDPARAMETER\",\"message\":\"Unknown parameter(s): excludeamc. Search parameters are strict so a typo can never silently return unfiltered results.\",\"docsurl\":\"https://api.tigzig.com/mf/v1/docs\",\"validparams\":\"q, limit, offset, amc, group, category, type, plan, option, txic, isin, active, stale, firstdatefrom, firstdateto, lastdatefrom, lastdateto, exclude\",\"examples\":[\"https://api.tigzig.com/mf/v1/search?q=absl%20value\",\"https://api.tigzig.com/mf/v1/search?category=small%20cap&plan=direct&option=growth\",\"https://api.tigzig.com/mf/v1/search?group=equity&firstdatefrom=2015-01-01&firstdateto=2020-12-31\"],\"didyoumean\":{\"excludeamc\":\"exclude\"}},\"help\":{\"catalog\":\"https://api.tigzig.com/mf/v1/\",\"docs\":\"https://www.tigzig.com/apis/mf-nav\"}}\n`\n\nThe error names the exact parameter it did not recognize, lists every valid one, and in this case guesses the one I meant, didyoumean pointing exclude_amc at exclude. category=value in that same call is a perfectly good filter on its own, and the response does not run that half of the call while dropping the other half unannounced. The whole call fails, with an explanation, so there is no version of this where a result set comes back missing a filter you thought had been applied, with nothing in the response to tell you so.\n\nSend the same filter parameter twice and the two rules differ by what kind of parameter it is. A filter that already supports commas as OR merges the repeats the same way.\n\n`\nGET https://api.tigzig.com/mf/v1/search?amc=birla&amc=hdfc\n`\n\nThis reads as amc=birla,hdfc, same 787 matches as writing it that way directly. q cannot merge, because it only ever takes one value, so sending it twice with two different values is rejected outright rather than keeping whichever one happened to arrive last with no indication of the other.\n\n`\nGET https://api.tigzig.com/mf/v1/search?q=hdfc&q=icici\n\n{\"detail\":\"Parameter(s) given multiple times with different values: q. For list filters (amc, category, ...) repeats merge as OR; this parameter takes one value.\",\"error\":{\"code\":\"REPEATED_PARAMETER\", ...}}\n`\n\nA few smaller edges\n\nA query of nothing but spaces used to pass the endpoint's own minimum length check and come back as a 200 with a fixed set of 30 results, no warning that the search behind it had run on an effectively empty string. That is closed now, whitespace collapses to nothing and gets rejected the same way a truly empty q does.\n\nlimit and offset are bounds checked cleanly, ask for limit=0 or limit=1000 and you get a 400 naming the field and the allowed range, 1 to 100, rather than a call that clamps the number to something else without telling you. Ask for a limit past 100 specifically and the error adds a pointer to the bulk file: /mf/v1/download?format=latest returns every scheme in one CSV or parquet file, so a caller trying to page through the whole universe one hundred rows at a time gets told there is a single GET that does it instead. A malformed date in firstdatefrom gets a 400 telling you to use YYYY-MM-DD.\n\nisin is the one filter that is not a substring match, it takes an exact ISIN only, comma separated for more than one. Every other text filter is contains matched.\n\nQuick tip: use last_date to find matured funds\n\nlastdate is easy to skim past as the throwaway half of the launch date pair, but it is the useful one for anyone studying matured funds, and that is not a small slice of this database. Roughly 29,000 of the more than 38,000 schemes are matured or merged, close to 77 percent of the whole universe. A query bounded by lastdatefrom and lastdate_to is how you find every scheme that stopped publishing inside a given window, fixed maturity plans included.\n\nLimitations and things to watch for\n\nEverything below is either a hard limit or a live gotcha to know about, none of it is a bug. Some of it repeats what is already covered above, kept here on purpose so this section stands on its own if it is the only part you read.\n\nNo boolean logic beyond AND and comma-OR\n\nFilters AND together automatically. A comma inside one filter is OR within that field. exclude subtracts at the end. That is the whole language, and there is nothing past it: no parentheses, no nested groups, no OR across two different fields, and no OR inside q itself, comma-OR only works on filters.\n\n`\ncategory=value,liquid&plan=direct\n`\n\nreads as category is Value Fund or Liquid Fund, AND plan is Direct. That is as deep as a single call goes.\n\nexclude only reads the scheme name and the AMC name\n\nexclude is one global NOT, it does not work per field. It drops a result if the term appears in the scheme's own name or its AMC's name, nowhere else. exclude=liquid removes any fund with \"Liquid\" in its name, which happens to catch most Liquid Fund category schemes, but it is matching name text rather than the category field. exclude=value also removes ICICI Prudential Nifty50 Value 20 Index Fund, a fund whose actual category is Index Funds, purely because the word \"Value\" sits in its name.\n\ncategory matches AMFI's raw text too\n\nCovered in full above, worth repeating here. category=index returns close to 1,300 schemes, the large majority genuinely carrying a categorysub of Index Funds, a few dozen pulled in because their raw AMFI category text also contains the word while their clean sub-category is something else, Debt Funds or Equity Funds. There is no filter today that reads the clean field alone, filter broad and check categorysub on the results.\n\nNo sort parameter, no typo tolerance\n\nResults come back in a fixed order, a relevance score on a q search, otherwise the order the database returns them in. There is no sort= or order_by=. Every text filter is an exact substring match, nothing fuzzy. category=lifecycle, written as one word, returns nothing, while category=life cycle, spaced the way AMFI actually spells it, returns the real matches. A misremembered filter value comes back as zero results, with no near-miss offered.\n\nPaging through more than 100 results\n\nlimit is how many rows come back in one call, capped at 100. offset is how many rows to skip before the response starts, offset=100 skips the first hundred and starts at row 101. Together they are the standard way to walk through a result set bigger than one call can hold.\n\namc=birla,hdfc from earlier in this post matches 787 schemes, more than eight pages at 100 a page. To read the whole set, call it repeatedly, moving offset forward by 100 each time, and stop once the rows collected reach the response's own total_matches:\n\n`\noffset=0&limit=100 -> rows 1-100\noffset=100&limit=100 -> rows 101-200\noffset=200&limit=100 -> rows 201-300\n...\noffset=700&limit=100 -> the last 87 rows\n`\n\nThat is for walking a filtered subset through search itself, a results page in an app, for instance. If the goal is every scheme in the database rather than a filtered slice, /mf/v1/download?format=latest is the shorter route, one call, no offset, nothing left to page through.\n\nThe nuance: a sweep spanning a sync\n\nPicture photographing a library shelf by shelf while the librarian is still shelving. Slot a new book into shelf 1 partway through and every book after it shifts one position over, the last book you photographed on shelf 3 reappears at the start of shelf 4. Pull a book out instead and everything shifts the other way, one book slips between two photographs and never gets photographed at all.\n\nThe database syncs three times a day. Page through a large browse with limit and offset, and a sync occurring mid-sweep can do either of these to your pagination, a row you already fetched reappears on a later page, or a row slides past your position and you never see it. The sweep finishes looking complete either way.\n\nThe nuance worth being precise about: this is not a risk on every sweep, or even most of them. It only shows up when a sync runs while a sweep is mid-flight, somewhere after the first page and before the last, and only if the row that moves happens to sit right at the page boundary the sweep is crossing at that exact moment. A sweep that finishes before the next sync, or starts after one, sees one clean, complete snapshot and never notices anything happened. Three syncs a day against a sweep that runs several minutes leaves a narrow, low-probability window for this, real but uncommon. For anything that needs to be exactly consistent regardless, a reconciliation, an audit, a research snapshot, use /mf/v1/download?format=latest instead of paging, it is written in one moment and cannot end up half old and half new.\n\nOne thing that is not a limitation\n\nEvaluation order is not something to learn here. Write the filters in any order and the result is identical, because the rule never changes: every filter ANDs together, a comma is OR within one field, exclude subtracts last. There is no WHERE-clause-style precedence to memorize the way there would be writing the equivalent query by hand in SQL.\n\nResources\n\nFor people\nAPI and MCP hub. The front door. Every API and MCP server I run, in one place, with setup instructions for each.\nMutual fund NAV API guide. The written walkthrough of this one API, with worked examples you can copy.\nMutual fund NAV API, interactive reference. The Swagger page. Every endpoint and parameter, and you can fire a live call from the browser to see the response shape.\nDemystifying Tigzig API errors. A longer guide to the common failures across the mutual fund, Tremor and Vigil APIs.\n\nFor your AI agents\n\nAI agents: this API is open, no auth. Start with the OpenAPI spec for this API at https://api.tigzig.com/mf/v1/openapi.json. The self-describing catalog at https://api.tigzig.com/mf/v1/ lists the endpoints and parameters. The RFC 9727 catalog at https://api.tigzig.com/.well-known/api-catalog is a site-level directory of every Tigzig API, go there only if you want a different API rather than more detail on this one. Site index for agents: https://www.tigzig.com/llms.txt`\n\nFound something that does not match what is written here, or a fund that will not turn up in search: there is a Report button in the MFPRO app header, and I read what comes through it.\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: MF NAV API docs, API + MCP hub, Mutual fund analytics hub\n\nMore posts: MFPRO Scheme Explorer is Live. Cross-Tab 38,000+ India Mutual Fund Schemes, and Watch the SEBI Re-Categorisation Happen in the Daily Files., SEBI Has Revised Mutual Fund Categorisation and Introduced a New Fund Type. The Major Changes, and What They Mean for MFPRO Users and Analysts., MFPRO Mutual Fund API Update. Three Things Your Calls Asked For, Now Built., Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users, Data Dictionary & Docs Live on MFPRO. Every Field Explained, 21-Metric Methodology, and How the India MF Universe Actually Looks."
},
{
"slug": "mfpro-search-upgrade-full-history-jul2026",
"title": "One Line of User Feedback, Fixed Within a Day. MFPRO Search Upgraded + Complete AMFI NAV History, 37M Rows, 2006 Onward.",
"tags": [
"mutual-funds"
],
"content": "One Line of User Feedback, Fixed Within a Day. MFPRO Search Upgraded + Complete AMFI NAV History, 37M Rows, 2006 Onward.\n\nSomeone typed one line into the Feedback box on MFPRO yesterday evening, \"ABSL Value fund and ICICI Pru Value Fund are missing\". Whoever you are, thank you, the fix went live within a day. With enhancements, including expanded search options.\n\nThe funds were in the database all along, the bug was in the search, and while at it the search got a proper upgrade, it understands ABSL and PPFAS and old fund names, and it ranks results by relevance rather than alphabet.\n\nThe second one - some days back users were hitting 404s on schemes that should have existed. I read every error in my logs, every day, and those 404s turned into a full rebuild, every NAV AMFI has ever published, 2006 onward. Now includes closed, defunct, interval & close ended funds also.\n\nNow from a single platform - run advanced analytics on the interface, access data via APIs, download the full database, 37M rows, in one shot, and even get Python codes to compute metrics offline.\n\nFor the methodology and the step by step validations, it is all in the Docs section on MFPRO.\n\ntigzig.com -> MFPRO\n\nMore on the search solution\n\nThe problem\n\nWe tend to type short names, the database stores long official ones. You type ABSL Value, the record says Aditya Birla Sun Life Value Fund, and a literal search finds nothing. Same story when a fund house renames itself, you remember Reliance Small Cap, the fund is now Nippon India Small Cap.\n\nThe fix has two parts\n\nOne, the search now keeps a list of these short forms and old names, and quietly expands them. Type ABSL, it searches Aditya Birla Sun Life. Type IDFC Flexi, it finds Bandhan Flexi Cap, the fund that renamed in 2023. Type PPFAS, it finds Parag Parikh. Seventeen mappings so far, each one added because a real search failed.\n\nTwo, results are scored instead of sorted alphabetically. A tighter match to what you typed scores higher, actively tracked funds score higher, so the fund you meant sits at the top instead of twenty index variants that happen to share a word.\n\nWill continue to modify the rules based on feedback and my own experience using the tool.\n\nThe validation ran at every stage rather than once at the end\n\nAt the pull: the full AMFI archive was downloaded and profiled offline first, nothing touched the main database at this stage.\n\nAt the parse: scheme category in the old archives lives in section header lines rather than on each record, so the parser carries that context through. Legacy category names were mapped to the current taxonomy through an explicit mapping I reviewed and approved, no silent guessing.\n\nAt the load: a pilot batch first, then date-ranged batches, and after every batch a cross-tab by year and scheme type, before plus added equals after, checked cell by cell before the next batch ran. The grand total closes to the row.\n\nExisting rows always win, and the nine conflicting overlaps found in 16.4 million rows are logged and documented.\n\nAfter the load: a separate multi-agent audit re-checked the whole chain, seven agents finding issues, seven adversarial agents verifying them. Zero duplicate scheme and date pairs across 37 million rows, zero orphans. Real items it caught were fixed the same day.\n\nAnd the source's own oddities, like zero-NAV rows in some 2006 era files, are published as filed in the docs rather than silently cleaned.\n\n![\"MFPRO](\"/images/blog/mfproupdate01feedback_fixes.png\")
\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, Quants Agent\n\nExplore: Mutual fund analytics hub, MF NAV API docs, Analysis hub\n\nMore posts: MFPRO Data & API Screen Live - Download the Whole India MF NAV Database in One Click. No Code, No API., Free Mutual Fund NAV API for India - Revamped. AI Ready. 18,000+ Schemes Since 2008 from AMFI., Now live - MFPRO v2 - Mutual Fund Analytics (India) - now with rolling returns, custom eval periods, multi-period & multi-instrument comparisons, Tigzig APIs - Reading the Logs, Shipping the Fixes. Status Page Live, R2 Downloads, Self-Correcting 4xxs."
},
{
"slug": "mfpro-sif-analytics-where-it-stands-aug2026",
"title": "Is SIF Analytics Coming to MFPRO? On the List, No Date.",
"tags": [
"mutual-funds"
],
"content": "Is SIF Analytics Coming to MFPRO? On the List, No Date.\n\nA few people have asked whether SIF analytics is coming to MFPRO. Here is where it stands: the plan is there, no date on it.\n\nSIF is SEBI's Specialised Investment Fund category, barely a year old and growing fast. But AMFI publishes SIF data on a different portal, and the daily NAV feed MFPRO syncs three times a day carries no SIF schemes at all.\n\nAnd by the way, the SIF ISINs look exactly like mutual fund ISINs, INF prefix and all, since SIF units sit under the AMC's mutual fund umbrella at the depository.\n\nSo, same identifier formats but different regulatory category and different source...and would need a new ingestion pipeline with its own validations.\n\nMFPRO itself has been live a few months and usage has picked up the past few weeks, in the app and through the API. I would rather let that settle and learn from it before adding SIF. Yes, on the list, no date.\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Mutual fund analytics hub, MF NAV API docs, API + MCP hub\n\nMore posts: SEBI Has Revised Mutual Fund Categorisation and Introduced a New Fund Type. The Major Changes, and What They Mean for MFPRO Users and Analysts., MFPRO Scheme Master: Every India Mutual Fund Scheme in One Small File, Free, Your Agents and Scripts Have Been Talking to Me. Here Is That Conversation in Plain English., MFPRO Scheme Explorer is Live. Cross-Tab 38,000+ India Mutual Fund Schemes, and Watch the SEBI Re-Categorisation Happen in the Daily Files."
},
{
"slug": "mfpro-v2-mutual-fund-analytics-rolling-returns",
"title": "Now live - MFPRO v2 - Mutual Fund Analytics (India) - now with rolling returns, custom eval periods, multi-period & multi-instrument comparisons",
"tags": [
"mutual-funds",
"portfolio-analytics"
],
"content": "Now live - MFPRO v2 - Mutual Fund Analytics (India) - now with rolling returns, custom eval periods, multi-period & multi-instrument comparisons\n\nMFPRO Rolling Returns\n\nFree to use. No login. No ads. Data Downloads. By Analyst. For Analysts.\n\nv1 just had portfolio comparison of one or more funds. New functionalities now added:\n\nReturns\nPoint-to-point returns - custom periods\nRolling return statistics: Average, Median, Min, Max across 1Y, 3Y, 5Y windows\nCustom evaluation periods for rolling returns\nCompare returns of multiple funds and indices on a single screen\nCoverage: 95 funds & 9 Indices\n\nDownloads & Validations\nRaw NAV data & Index data as CSV for independent offline verification\nPre-built Excel templates with CAGR & Rolling return formulas for cross-checking\nRolling return validation: e.g for BSE 100 - 7,485 values verified within 0.50 bps of manual calculation\nStep-by-step guide on how to replicate every number independently\n\nPortfolio Holdings\nCompare holdings of one or more funds side by side across two time periods (existing)\nCompare fund holdings against index benchmarks (Nifty 50, Nifty 100, Nifty 500, Nifty Midcap)\nMulti-period comparison - track holdings changes across up to 5 time periods\nCoverage: 24 funds across 6 categories: Large Cap, Mid Cap, Small Cap, Flexi Cap, Focused, Value\n\nIndices\nView constituents and weights of Nifty indices (Nifty 50, 100, 500, Midcap, Midcap 50)\nCompare two indices side by side - see overlapping and unique stocks\n\nData\nPortfolio composition from MF Monthly Disclosure files\nNAV & Indices from Jan 2013 & sourced from mfapi.in\nAutomated daily sync - data updated twice daily\nView daily update status on app\n\nDocs\nDocs with detailed explanation of methodology with examples.\n\nLive now at: mfpro.tigzig.com\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, Midcap Dashboard, QRep - Security Reports\n\nExplore: Mutual fund tools hub, Market tools hub, Guides and releases\n\nMore posts: Rolling Returns: Why CAGR Alone Can Mislead You (And What To Use Instead), Releasing MDRIFT - Mutual Fund Composition & Drift Analytics Tool, That 9X return from Nifty Midcap is irrelevant if you couldn't survive the 73% of time it was in drawdown, Analysis-as-App: Inside India’s Top Midcap Funds: Buys, Sells, Entries and Exits. Interactive Dashboard Release (Analysis-as-App)"
},
{
"slug": "mfs-collapse-cockroaches-private-credit-may2026",
"title": "Dimon Was Right. More Cockroaches Arrive. UK Lender MFS Collapses - $2.5B Across 12+ Global Banks.",
"tags": [
"portfolio-quants"
],
"content": "Dimon Was Right. More Cockroaches Arrive. UK Lender MFS Collapses - $2.5B Across 12+ Global Banks.\n\nPublished: 2026-05-18\n\nUK lender MFS collapsed in February with $2.5B exposure across 12+ global banks - Barclays $308M, HSBC $400M, Atlas/Apollo $508M, Santander $267M, Elliott $254M, Wells Fargo $182M. £2.4B loan book, fraud and double-pledging allegations. £230M verified collateral against £1.16B owed (80% shortfall). Affiliated lending chain: HSBC > Apollo Atlas > Zircon Bridging > MFS.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/mfs-collapse-cockroaches-private-credit-may2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/mfs-collapse-cockroaches-private-credit-may2026 with header Accept: text/markdown\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nTIGZIG · Private Credit Note · 18 May 2026\n\n\"When you see one cockroach, there are probably more.\"\n— Jamie Dimon, October 2025\n\nMFS · £2.4B UK lender → fraud, gone\n\nWhat happened. UK specialty-property lender, Mayfair. February 25, 2026 insolvency. Same properties double-pledged across multiple banks. £230M verified collateral vs £1.16B owed — an 80% shortfall.\n\nDisclosed exposure by lender\n\n| Lender | Exposure |\n|---|---|\n| Atlas (Apollo) | $508M |\n| HSBC | $400M |\n| Barclays | $308M |\n| Santander | $267M |\n| Elliott | $254M |\n| Wells Fargo | $182M |\n| Jefferies | $135M |\n| SMBC | $127M |\n\nPlus Avenue Capital $124M, Castlelake $89M, IKB $39M, Macquarie <$63M.\n\nSources: \"Why a small UK lender has major U.S. credit firms on edge\", CNBC, 18 May 2026; \"The opaque private-lending deals that left HSBC with a $400M hole\", WSJ, 5 May 2026. Totals not strictly additive; back-leverage chains overlap (e.g. HSBC routed via Apollo's Atlas).\n\nThe pattern, stacked\nFrauds. First Brands (Sep 25). Tricolor (Sep 25). DOJ probes both.\nOvervaluations. 0.78x median BDC market price / NAV.\nExposure. Up to approximately $1.57T US banks. $2.4T US life insurance.\nRedemptions. Unprecedented surge in redemptions and gatings.\n\nDefault outlook\n\n\"We expect direct lending default rates to reach 8%, approaching Covid peak levels.\"\n— Morgan Stanley (Joyce Jiang · CNBC, March 17, 2026)\n\nFurther reading\n\nAnalysis: tigzig.com/analysis\nRed Flag in US Life Insurance\nPrivate Credit, Ghosts of 2008\nUS Banks NDFI Exposure\n\nTools: tremor.tigzig.com\nUS Banks and Credit Unions\nUS Life Insurance\nGlobal Macro Indicators\n\nAmar Harolikar · Decision Sciences & Applied AI · tigzig.com\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: Private Credit Stress Hub, Quant Apps hub, Analysis deck hub\n\nMore posts: Private Credit Default Rate Hits a Record 6.0% (2026): A Market for Lemons? Banks and Insurers Deep In It., Banks Holding the Line.. But the 'Lemons' Sit Off the Chart. In Private Credit, Credit Unions, and Non-Bank Consumer Lenders., Private Credit - The $2.7 Trillion Shadow Lending Market Is Showing Cracks, The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top., Private Credit and BDC Stress. Discounts at 13-Year Lows. Withdrawals Past the Gate. Regulators Now Arriving."
},
{
"slug": "mistakes-i-made-building-text-to-sql-agents-live-projects-2025-learnings",
"title": "Mistakes I Made Building Text-to-SQL Agents in Live Projects. My 2025 Learnings",
"tags": [
"text-to-sql",
"database-ai"
],
"content": "Mistakes I Made Building Text-to-SQL Agents in Live Projects. My 2025 Learnings\n\nI've been setting up Text-to-SQL apps and agents for over a year now. Multi-user systems. Live databases. Made many mistakes. Here's what broke, what it cost, and how I fixed it.\n\nSTRATEGIC & ARCHITECTURE\n\n▸ Built for analytics when main use was operations\n\nDesigned for analysts running complex queries. Actual users: customer service, finance & operations teams pulling raw data for reconciliations and responses.\n\nMost queries are straightforward - ID lookups, transaction history, ad-hoc filters. Don't over-engineer for complexity that doesn't exist.\n\n▸ Using multi-step agents where simple one would do\n\nBuilt a reasoning-heavy agent: plan → execute → validate. Cost went from <$1 per 100 queries to $20+. Queries took 1-3 minutes. Single-step agents work for 80% of SQL tasks. Use multi-step only when genuinely required.\n\nDATA & CONTEXT SHARING\n\n▸ Didn't share schema with the agent\n\nUser asks query. Agent runs blind - wrong joins, hallucinated columns, missed filters.\n\nAlways include: table names, columns, data types, primary/foreign keys, relationships.\n\n▸ Did not provide business context\n\nAgent doesn't know what fields mean, how they're computed, what edge cases exist.\n\nShare: variable definitions, computation logic, common query patterns, odd values to expect, business rules. Schema alone isn't enough.\n\n▸ Used plain text dumps for schema\n\nDumped schema as big text blocks adding things randomly. Agent got confused.\n\nUse YAML - is human-readable, cleaner hierarchy.\n\n▸ Didn't share categorical variable distributions\n\nAgent didn't know valid values for status fields, product categories, regions. Made up values. Spelling inconsistencies broke queries.\n\nShare distributions for cardinal variables upfront. Agent needs to know: what values exist, common misspellings, edge cases. Gives it X-ray vision into the data. Basically share a condensed EDA output with the agent.\n\nAGENT INSTRUCTIONS & CONSTRAINTS\n\n▸ No LIMIT clause rules\n\nAgent would download full datasets for 'exploration'. 1M rows pulled for simple queries.\n\nRule I added: Always LIMIT 100 on raw SELECT queries. No LIMIT on aggregations (COUNT, SUM, GROUP BY). Explicitly tell the agent when and when not to use LIMIT. For stricter control, apply at DB level (if supported) or API layer. This is a tricky one though.\n\n▸ Unlimited query execution\n\nAgent would run say 10 SQL queries for a simple question. Cost and latency go up.\n\nCap it. Simple agent: 2-3 queries max. Complex reasoning agent: 7-10. Specify execution limit upfront based on use case.\n\n▸ No debug protocol\n\nQuery failed. Agent didn't self-diagnose.\n\nGive the agent a debug protocol: That it should try and debug iteratively in case of errors, say up to additional 5 queries.\n\n▸ Assumed agents write perfect SQL\n\nThey don't. Division by zero errors is very common.\n\nTell the agent: wrap divisors with NULLIF(x, 0). Use COALESCE() for fallback values. Identify common error patterns and suggest fixes upfront.\n\nAGENT BEHAVIOR REALITY\n\n▸ Assumed agents follow rules 100%\n\nSet LIMIT rules, query caps, table restrictions. Agents still break them occasionally. Non-deterministic behavior.\n\nYou will never get 100% compliance. You can add additional checks at DB level or API level. But that's the nature of the beast. See what you can live with.\n\nMODEL SELECTION & COST OPTIMIZATION\n\n▸ Used expensive models everywhere\n\nUsed to use GPT-4.1 for everything. 5-10x cost vs cheaper models.\n\nGPT-4.1-mini and even GPT-4o-mini great for routine tasks. Reserve GPT-4.1 / 5.1 / 5.2 for harder tasks.\n\n▸ Upgraded to latest model immediately\n\nSwitched from GPT-4.1 to GPT-5 the day it launched. Cost blew up - reasoning tokens added massive overhead.\n\nLet models stabilize. Test cost and performance before upgrading. Don't chase 'latest' by default.\n\n▸ Didn't specify reasoning parameters\n\nMany models have reasoning controls. Didn't use them. Agent applied full reasoning to simple queries. For simple tasks, set reasoning to zero or minimal. Saves tokens and latency. Only enable reasoning when genuinely needed.\n\n▸ Didn't monitor cache hits\n\nSystem instructions and schemas ran 20K+ tokens. No cache hits = repeated full charge.\n\nMonitor cache hit rates in logs. Key rules: exact prompt match (byte-for-byte), no dynamic content at start, proper cache configuration. This is a complex area with major cost impact - detailed cache optimization guide coming soon.\n\nSECURITY & ACCESS CONTROL\n\n▸ Provided write access inadvertently\n\nUser truncated live tables. My fault - the test ID write permissions went to production.\n\nFix: separate DB users with fine-grained permissions.\n\n▸ No CREATE table functionality\n\nUser pulls 5,000 records. Can't print it. Can't export it.\n\nAdded: write access to 'workdir' schema. Added export functionality. Gave users control over larger result sets.\n\nPERFORMANCE & LIMITS\n\n▸ No rate limits\n\nOne user could trigger 100+ queries in a minute.\n\nRate limits at API level (SlowAPI for FastAPI). Rate limits at agent level.\n\n▸ Async in name only\n\nWrote async def everywhere. One blocking file read killed concurrency for all users.\n\nAudit for hidden blocking calls. Wrap blocking DB operations in thread pools. Replace blocking file reads. Async must be end-to-end or it doesn't work.\n\n▸ Didn't test edge cases\n\nWorked fine with clean data. Broke with null values, special characters in column names, mixed delimiters, inconsistent date formats. Push it to limits. Test with messy real-world data. Observe failures. Adjust instructions.\n\n▸ Temporary tables without cleanup\n\nAllowed agent to create intermediate tables. Never dropped them. Database bloated over time.\n\nHave agent use clear naming (e.g. tempagentxxxx) and purge regularly.\n\nResult\n\nCost per 100 queries: $20+ → under $2 for most tasks\nLatency: 2+ minutes → 10-30 seconds\nError rate: ~30% → under 5%\nUser satisfaction: measurable improvement\n\nTest it\ntigzig.com/analyzer - live Text-to-SQL app with sample datasets up to 1.6GB. No login required. Open Source\ntigzig.com/database-landing - detailed field guides & costing for deploying AI on databases\ntigzig.com/blog - posts and guides on AI for Analytics\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, BRIQ, Database Query Tool\n\nExplore: Database AI Apps, APIs and MCP, Blog\n\nMore posts: LLM Costing for Database AI Apps. Live Experience. Live App. Open Source, Large File Upload for Database AI Text-to-SQL Apps: A Practical Guide, AI for Databases: Field Guide, Live Apps & Lessons, Can an AI SQL Agent build a weighted scoring system from scratch?"
},
{
"slug": "monthly-mf-portfolio-files-hours-wasted-re-formatting-here-s-a-tool-that-fixes-it",
"title": "Monthly MF portfolio files = hours wasted re-formatting. Here’s a tool that fixes it",
"tags": [
"mutual-funds",
"converters-tools"
],
"content": "Monthly MF portfolio files = hours wasted re-formatting. Here's a tool that fixes it\n\nWhy This Exists\n\nIf you track Indian MF portfolios, you know the drill. Monthly disclosure Excel files = same info, different layouts. Spending time re-structuring is dead time.\n\nWhat You Get\nA converter utility built for analysts.\nDrop in raw files.\nGet back clean CSVs with ISIN mapping + standardized names.\nPlug straight into your analysis workflow\n\nHow It Works\nAI-powered schema detection: Automatically figures out data layout\nCross-validation: Runs multiple models to check totals.\nOverride: When in doubt, enter the schema manually - tool re-runs instantly.\n\nExtra Utilities\nAppend multiple funds → one file.\nTranspose holdings → side-by-side view.\nLogs + validation table → quick audit trail.\n\nWhat Next - Analysis Options\nDo it your way → pivots, Jupyter, Colab.\nUse the AI Analyst SQL Agent at www.tigzig.com/analyzer → Choose File Load to a temporary DB → analyze with Claude Sonnet 4 / DeepSeek / Qwen3 Max.\nUpload to ChatGPT (Go/Plus). Have it load into python/ pandas and analyze.\n\nBottom line: A single, focused utility. Converts messy disclosures → clean, structured data. From there, analysis is up to you. Usage guide + build guide included → hit 'Docs' on top.\n\nLive at\n\nwww.tigzig.com/mf-files-ai\n\nResources for Analysts\nRBICC Tool: If you also work with RBI Monthly ATM/ POS/ Cards, this Excel to CSV converter might be useful - www.tigzig.com/rbi-cards\nFull Suite: All 25+ open-source analytics apps are here → www.tigzig.com\n\n\nRelated\n\nTools: MF File Converter, MFPRO - Mutual Fund Analytics, DATS-4 Database AI Suite\n\nExplore: Mutual Fund Analytics hub, Database AI Apps, Market and Macro Tools\n\nMore posts: AI automation micro-app: MF Portfolio Files Processor. Live app. Open source., Releasing MDRIFT - Mutual Fund Composition & Drift Analytics Tool, Free Mutual Fund NAV API for India (AMFI) - No Key, Download 20.5M+ NAVs in One Call"
},
{
"slug": "movie-similarity-engine-sql-jaccard-duckdb",
"title": "How I Built a Sub-Second Movie Similarity Engine With a 10-Line SQL Query",
"tags": [
"duckdb",
"database-ai"
],
"content": "How I Built a Sub-Second Movie Similarity Engine With a 10-Line SQL Query\n\nOn my IMDb dashboard (tigzig.com/movie-explorer), you can click a lightning icon next to any movie and get a list of similar movies. Match percentages, shared factors, the whole thing. It runs in under a second.\n\nNo ML - just SQL\n\nThe Problem\n\nIMDb has 12 million titles and 97 million person-to-title credit records. I wanted a 'movies like this' feature that could compare any movie against all others and return results fast enough to feel instant. The constraint was that I only had public metadata - no user ratings or watch history. Just titles, genres, cast, crew, year, and runtime.\n\nBrainstormed with Claude Code on approaches - from ML-based recommender systems to simpler methods - looking at trade-offs, deployment constraints etc. Here's what I went with.\n\nThe Approach: Jaccard Similarity With Weighted Tokens\n\nJaccard similarity is simple. Take two sets. Count what they share. Divide by the total unique items across both sets. Two movies that share 5 out of 20 total unique factors get a 25% match score.\n\nThe trick is what you put in those sets. I capture each movie's attributes as a list of tokens - text labels that describe its properties. A movie might have tokens like genre:Drama, director:nm0000229, star:nm0000151, decade:1990s, runtime:standard, rating_band:excellent.\n\nThe problem is that standard Jaccard treats every token equally. Sharing a lead actor should matter more than sharing a runtime category. But Jaccard has no built-in weighting mechanism.\n\nThe workaround is token duplication. I repeat important tokens multiple times in the list. A top-billed star gets about 10 tokens - star:, lead:, and actor: variants of their ID, repeated. A director gets about 5 tokens. A writer gets 1. Runtime gets 1. Genre gets 1. When two movies share a star actor, those ~10 duplicate tokens all appear in the intersection, pushing the Jaccard score up far more than sharing a single runtime:medium token would. The weighting is baked into the data at table creation time - the similarity query itself is completely unweighted. It just counts matching tokens versus total tokens. The weighting already happened.\n\nThe factors I settled on: genre, directors, actors (with star, lead, and supporting distinctions), writers, producers, decade, runtime category, and rating band.\n\nThe Implementation\n\nI cannot run this comparison on raw data. The cast and crew information lives in titleprincipals - 97 million rows. Genres and runtime are in titlebasics - 12 million rows. Ratings are in title_ratings - 1.6 million rows. To build a token set for one movie, you need to join across all three tables. To compare one movie against all others on the fly, you'd be hitting that 97 million row table repeatedly.\n\nWhat I did was build a pre-computed table called movie_tokens. One row per movie, with all similarity factors already assembled from those three source tables and encoded as a DuckDB list (array) column.\n\nThe key design choice: I don't tokenize all 12 million titles\n\nI filter to titleType = 'movie' (drops TV shows, shorts, episodes, etc.) and then to movies with 10,000+ votes. That takes 12 million titles down to about 12,000 movies. These are movies that enough people have actually watched and rated. Nobody needs a similarity match for a zero-vote short film from 1927. This filter is what makes the whole thing fast - the similarity query compares against 12,000 rows, not 12 million.\n\nSo instead of joining across 97 million + 12 million + 1.6 million rows at query time, I join them once during the data pipeline and store the result. The movie_tokens table is about 10MB. Each row has a tconst, title, year, rating, votes, genres, and a tokens array.\n\nThe query that powers the whole feature\n\n``sql\nWITH source AS (\n SELECT tokens FROM movie_tokens WHERE tconst = 'tt0111161'\n)\nSELECT\n m.tconst, m.title, m.year, m.rating, m.votes, m.genres,\n LEN(list_intersect(m.tokens, s.tokens)) * 1.0 /\n LEN(listdistinct(listconcat(m.tokens, s.tokens))) as jaccard_score,\n listintersect(m.tokens, s.tokens) as matchingtokens\nFROM movie_tokens m, source s\nWHERE m.tconst <> 'tt0111161'\n AND LEN(list_intersect(m.tokens, s.tokens)) > 0\nORDER BY jaccard_score DESC, m.votes DESC\nLIMIT 30\n``\n\nBut building the pre-computed table is only half the problem\n\nWhen a user clicks that lightning icon, the query still has to take one movie's token list and compare it against all 12,000 other movies in real time. That's 12,000 set comparisons - find the intersection, find the union, compute the ratio - all in one query.\n\nThis is where DuckDB earns its keep. The CTE (WITH clause) grabs the source movie's tokens. Then the main query cross-joins that against every other row in movietokens. For each pair, listintersect finds the shared tokens, listconcat + listdistinct gives the union, and the division produces the Jaccard score. The matching_tokens column returns exactly which tokens matched - those are the tags you see in the \"Why Similar\" column on the UI.\n\nDuckDB is built for this kind of columnar, analytical workload. It processes all 12,000 comparisons - each involving array intersection and union operations - in under a second. It's not doing 12,000 separate queries. It's one vectorized scan across the entire table. That's the difference between an analytical database like DuckDB and a row-oriented database.\n\nThe backend cache means repeated lookups for the same movie are instant after the first hit.\n\nJaccard on token lists is not the most sophisticated similarity algorithm\n\nBut it has properties I care about: it's explainable (you can show exactly which factors matched), it runs in pure SQL (no external libraries or services), and it's fast enough for a real-time feature.\n\nThe matching_tokens field is what makes this useful. When a user sees '24% match' next to a movie, they can also see Actor, 1990s, Drama, Lead, Star - the actual reasons. That transparency matters more than a slightly better algorithm that feels like a black box.\n\nThe full implementation is open source. Frontend code, backend code, the token generation query, the similarity SQL - all in the repos linked on the dashboard. The data pipeline docs walk through exactly how the movie_tokens table is built.\nLive app: tigzig.com/movie-explorer (go to Explore tab, click any lightning icon)\nFrontend: github.com/amararun/shared-imdb-dashboards\nBackend: github.com/amararun/shared-duckdb-dashboards-backend\n\n\nRelated\n\nTools: Movie Explorer, BRIQ, DATS-4 Database AI Suite\n\nExplore: Database AI Apps, Browse all apps, Guides and build notes\n\nMore posts: CinePro - 230M Rows, 16GB Database, Instant Queries with DuckDB, From 12 second queries to under 1s: Optimizing a 230 Million Row Dashboard - 14 Bottlenecks I Had to Fix, Architecture & Setup for a Dashboard with Hundreds of Millions of Records - Powered by DuckDB, BRIQ App: DuckDB AI in Browser - 500MB Files, 4M+ Records, No Database Setup, DuckDB isn't just fast SQL. It's Python, SQL and compression all in one box."
},
{
"slug": "multi-agents-sequential-reasoning-connect-database-o3-mini-deepseek-r1-flash-2-0-flowise",
"title": "Multi-Agents (Sequential) with Reasoning – Connect to any database - o3-mini / Deepseek-R1 / Flash-2.0. Built with Flowise.",
"tags": [
"database-ai"
],
"content": "Multi-Agents (Sequential) with Reasoning – Connect to any database - o3-mini / Deepseek-R1 / Flash-2.0. Built with Flowise.\n\nBuilt with Flowise. Leverage as a reusable component\n\nWith agent templates, tool schemas and source codes.\n\n6 database agent options\n\nThe Analyzer micro-app on REX-3 (tigzig.com) offers 6 agent options, all capable of:\nconnecting to any database (with credentials)\nnatural-language-to-SQL\nnatural-language to python (for charts & stats)\n\nApp also allows on-the-fly Postgres database for file uploads and temp use\n\nAI backend is on Flowise AI. Leverage as a reusable component:\ndirectly via Flowise's native UI (Click → Share Chatbot → Live)\nplugged into any UI via API (as in REX-3)\nintegrated into automated workflows via API\n\nEasy to setup, modify and maintain.\n\nStandalone General Analyst\n\nSingle-agent setup for regular queries & simple analysis. Uses o3-mini; for complex cases, I use gpt-4o\n\nMulti-Agents (technically sequential agents in Flowise, Backend: LangGraph)\n\nThree-agent flow: Gemini-Flash-2 / o3-mini / Deepseek-R1. Agent flow:\nRouter Agent → Determines if a query goes to General or Advanced Analyst\nGeneral Analyst → Same as standalone version above\nAdvanced Analyst →\nStep 1: Reasoning LLM (o3-mini / Flash-2 / Deepseek-R1) generates analysis plan & SQL\nStep 2: gpt-4o agent verifies query accuracy, executes, and returns results\n\nMulti (Sequential) Agents with Additional Debugging\n\nStill in testing. Similar to sequential agents but with:\none has and additional error-checking step\nother is a debugger agent for routing based on user request\n\nTake them for a spin\n\nFeel free to test out with different data/query complexity/agent choices-no API key needed. Connect to your database or use the temporary Postgres DB (created on-the-fly for file uploads). Upload your own files or just hit the 'Sample' button for sample file upload.\n\nSource Codes/ Resources\nFlowise agent templates & tool schemas\n\nAll schemas + source codes → www.tigzig.com (Help & Build section)\nDetailed Video Guide – Demo & Build Walkthrough\n\nCovers: Agent architecture, SQL errors, API failures, error handling, cost, performance, quality, latency\nTop Flowise Resource\n\nLeon van Zyl's YouTube → @leonvanzy (best guide out there)\n\nExplore 15+ open-source AI analytics apps at tigzig.com with natural language to SQL, python charts, stats, database connectivity and more. Free access, source codes and build guides\nAI Co-Analyst Advanced Analytics App: https://lnkd.in/g8xPQvb8\nAI Automation Micro App: https://lnkd.in/gxVbCdpv\nRealtime SQL Voice AI: https://lnkd.in/g6cT8ZSM https://lnkd.in/gCT9J_K4\nConnect ChatGPT to Supabase in 10 minutes: https://lnkd.in/g2A9h8f2\n\n\nRelated\n\nTools: DATS-4 Database AI Suite\n\nExplore: Database AI Apps hub, APIs and MCP servers, Browse all apps\n\nMore posts: AI Driven Advanced Analytics. Reasoning based Sequential Agents. Connect to any database — o3-mini/deepseek-r1 / gemini-flash-2.0., Quick Deploy Advanced Analysis Multi-Agent with Flowise, Flowise is my goto platform for GenAI and LLM apps, Mistakes I Made Building Text-to-SQL Agents in Live Projects. My 2025 Learnings"
},
{
"slug": "mutual-fund-analysis-custom-gpt-python-multiple-excel",
"title": "Mutual Fund Allocation Analysis with GPT Power Tools. Custom GPT. Custom Python Code. Multiple Excels.",
"tags": [
"mutual-funds",
"custom-gpt"
],
"content": "Mutual Fund Allocation Analysis with GPT Power Tools. Custom GPT. Custom Python Code. Multiple Excels.\n\nLive use case of having the GPT use a custom code that ensures consistent performance, rather than relying on it's adhoc different approach each time. A GPT Power Tool.\n\nIn my previous videos and posts, I have shown how to use GPT Plus to run machine learning (https://rb.gy/o39bvm), LlamaParse complex PDFs to analyze 10K / Annual Reports(deployed at https://www.tigzig.com/). As well as a LLM App,YFIN Bot that will pull multiple-year financials from Yahoo Finance, now live on public internet (https://www.tigzig.com/yfin-bot) as well as a custom GPT (https://rb.gy/fy8lkk)\n\nAll apps, including this one, fully coded by ChatGPT / Gemini (wherever coding is required)\n\nCurrent use case: (a) track changes in equity holdings of a mutual fund over two or more time periods. (b) compare portfolio holdings of two or more different mutual funds. To be done repetitively. Can handle upto 10 excel files, each in different format (GPT Plus limit)\n\nSolution: Custom GPT 'Mutual Funds Portfolio Analyzer' (https://rb.gy/nj51lf). Live now. Feel free to use. It has regular instructions along with a custom Python code generated by GPT itself, to serve as a guideline. I was getting inconsistent results with just prompts and instructions. So now I am having the GPT use a coding approach that ensures consistent performance, rather than relying on it's adhoc different approach each time. A GPT Power Tool.\n\nHow to Use: Upload the excels, share a few details and that's it. Have GPT analyze it or analyze offline, pushing the appended data into a Pivot Table. [See document for details]\n\nWhere do I get the portfolio allocation files? Available on individual MF websites.\n\nCan I replicate this for my excel processing use case, nothing to do with equities? Yes. Just replicate the GPT and customize your instructions. Easy to do. No coding required. Only one piece of GPT auto generated code to be copy pasted during setup. (see document)\n\nCan I deploy this outside of GPT store, on public internet or my corporate intranet Yes, you can. See my YFIN bot for pulling multi-year financial statements from Yahoo Finance (https://rb.gy/ca2k5p). It's deployed inside GPT store as well as on the public internet and uses GPT 3.5. However for real-world use, anything requiring GPT4/Code Interpreter has significant cost considerations. Currently exploring lower cost custom build option. Watch this space for more.\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, MF Holdings Processor (GPT), MF Excel to CSV (GPT)\n\nExplore: Mutual Fund Analytics hub, Custom GPTs, Quant Apps hub\n\nMore posts: Mutual Fund Portfolio Analysis with ChatGPT: Merging and analyzing across multiple excel files, Monthly MF portfolio files = hours wasted re-formatting. Here’s a tool that fixes it, AI automation micro-app: MF Portfolio Files Processor. Live app. Open source."
},
{
"slug": "ndfi-analytics-us-bank-exposure-tool",
"title": "NDFI Analytics - US Bank Exposure to Non-Bank Lenders",
"tags": [
"portfolio-quants"
],
"content": "NDFI Analytics - US Bank Exposure to Non-Bank Lenders\n\nPublished: 2026-03-31\n\nInteractive NDFI analytics tool covering US bank exposure to non-bank lenders. FFIEC Call Reports for 629 banks across 5 quarters. Stress scenarios, CET1 concentration, ACL coverage, nonaccrual rates, commitments. Validated against FDIC BankFind API, FRED Series, FDIC Quarterly Banking Profile. Full data download.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/ndfi-analytics-us-bank-exposure-tool\nMarkdown of the HTML page: send GET https://tigzig.com/post/ndfi-analytics-us-bank-exposure-tool with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/NDFCTOOLMARCH2026.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nNDFI Analytics ToolUS Banks Lending to Non-Depository Financial Institutions (NDFI)\n\nOnline Interactive Tool\n\nStress scenarios to capital impactConcentration risk across 629 banksCET1 exposure & ACL coverage\n\nBank-level drill downs and full data download\n\nSource: FFIEC Call Reports\n\nValidated against FDIC BankFind API, FRED Series and FDIC Quarterly Banking Profile\n\nAmar Harolikar, ACA\n\nTigZig.com\n\n629 NDFI Lending Banks. Representing 88% of banking system assets (FDIC Insured banks). Quarterlies from Q4’24 to Q4’25 .\n\nThe Tool\n\nAmar Harolikar. Decision Sciences & Applied AI tremor.tigzig.com\n\nLive on tremor.tigzig.com ➔ US NDFIBuilt by Analyst. For Analysts\n\nThe questions you can answer\n\nAmar Harolikar. Decision Sciences & Applied AI tremor.tigzig.com\n\nWhich banks above $100B have NDFI exposure exceeding 150% of CET1?\n\nHow has NDFI concentration changed over the past 5 quarters for a specific bank?\n\nAt a 15% NDFI loss rate, which banks would see losses breach their entire ACL and start eroding CET1 capital?\n\nFor a specific bank, what percentage of its CET1 is exposed to NDFI?\n\nIf NDFI balances grow another 35%, what does the system-level NDFI/CET1 ratio look like?\n\nWhat is the commitment-to-balance ratio by NDFI subcategory, and which banks have the highest undrawn commitments?\n\nHow much of the NDFI portfolio is in Business Credit vs PE Funds vs Mortgage at any given bank?\n\nSections\n\nAmar Harolikar. Decision Sciences & Applied AI tremor.tigzig.com\n\nCoverage: which banks covered and how much of banking system do they represent?\n\nAnalysis: stress test scenarios and capital impact\n\nBalances : NDFI balances by subcategory and bank size\n\nCommitments : by subcategory and bank size\n\nExposure: capital adequacy and loss reserve ratios and breakdowns\n\nCredit Quality: non accrual rates and breaks.\n\nBank : Bank NDFI profile\n\nExplorer: data explorer to cut data as you want\n\nMethodology: detailed process followed\n\nValidation : validation process and results\n\nAnalysis\n\nAmar Harolikar. Decision Sciences & Applied AI tremor.tigzig.com\n\nAnalysis with stress scenarios, exposure analysis and potential for systemic risk.\n\nGet to insights fast\n\nAmar Harolikar. Decision Sciences & Applied AI tremor.tigzig.com\n\nApply filters, see % distros by numbers and volume, click on any bar to see bank list ... cick on bank to go to bank page…\n\nExplore and cut data\n\nAmar Harolikar. Decision Sciences & Applied AI tremor.tigzig.com\n\nData explorer view – apply multiple filters to data, select columns from 55+\n\nDownload data and analyze offline\n\nAmar Harolikar. Decision Sciences & Applied AI tremor.tigzig.com\n\nFull data download provided\n\nMethodology\n\nAmar Harolikar. Decision Sciences & Applied AI tremor.tigzig.com\n\nDetailed methodology including the MDRM codes used for variables and formulas\n\nValidations\n\nAmar Harolikar. Decision Sciences & Applied AI tremor.tigzig.com\n\n8 Levels of validations from internal consistency to total tally with external number\n\nTotals validated against FDIC Quarterly reports, FDIC Bank Find API & FRED Series\n\nValidation results shared with variances and reason for variance where identified.\n\nLive on TREMOR tremor.tigzig.com\n\nMacro Early Warning Signals\n\nPart of the Tigzig analytics suite - 40+ tools at tigzig.com\n\nAmar Harolikar. Decision Sciences & Applied AI tremor.tigzig.com\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QRep - Security Reports, Quants Agent\n\nExplore: Private Credit & NDFI Tracker, Quant Apps, Analysis\n\nMore posts: Banks Keep Funding the Non-Banks. $1.65T in Q1 2026, Up 42% in Five Quarters. US Banks NDFI Update., US Bank Exposure to Non-Bank Lenders - Now in an Interactive Analytics Tool, US Banks & Non-Bank Lending - How Deep is the Exposure? Is there a Systemic Risk?, Bank Loss Rates Look Calm. The Stuff Underneath - Consumer, Private Credit, Non-Banks - Says Otherwise., Releasing US-CREDIT. For Analysts Tracking US Banking. Interactive Tool Covering FDIC SDI and NY Fed CCP."
},
{
"slug": "new-open-source-tool-mutual-funds-holdings-analyzer-python-in-excel-xlwings-lite-now-live",
"title": "New Open Source Tool. Mutual Funds Holdings Analyzer. Python in Excel (xlwings Lite). Now Live.",
"tags": [
"mutual-funds",
"xlwings-lite",
"python-in-excel"
],
"content": "New Open Source Tool. Mutual Funds Holdings Analyzer. Python in Excel (xlwings Lite). Now Live.\n\nIt's a specific solution to a recurring problem: standardizing, consolidating and comparing mutual fund portfolio data from the monthly statutory disclosures (India Market). The pipeline takes monthly portfolio disclosure files from mutual funds as inputs and outputs formatted summary with key indicators and charts.\n\nName standardization and ISIN merging\n\nThe workflow standardizes multiple security names for same ISIN and allows for merging security names for corporate action via a human-in-the-loop intervention step.\n\nAI-Powered Data Conversion\n\nRaw portfolio disclosure files (Excel) are fed into a separate converter tool (HTML-JS). It uses AI schema detection to identify structure, extract the data, and standardize it into a single text file. This is the input for the xlwings Lite App.\n\nxlwings Lite automation\n\nImport and parse data with a file picker (VBA), with automatic delimiter detection. Run Stage 1 to execute core Python script. It loads the data, auto detects the two reporting periods, and performs the initial analysis. Its primary job is to generate data quality reports that flag all naming and mapping inconsistencies. After a human review and modification of the ISIN_Mapping file, Run Stage 2 - the script re-runs the final aggregation using the cleaned mapping file, with final summary and charts.\n\nHuman-in-the-Loop\n\nThe process separates automated analysis from manual intervention. The Stage 1 reports are designed to expose data quality issues. Quality is enforced by the manual review before the final report is generated.\n\nAI Coder used for the app\n\nGemini CLI\n\nxlwings Lite MF Holdings Analyzer\n\nApp & Docs\n\nResources\n\nMutual Funds Holdings Analyzer\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, MF File Converter, AI Schema Detection\n\nExplore: Mutual Funds hub, xlwings Lite Practice Lab, AI in Excel hub\n\nMore posts: How I Identify and Map Every Holding — The ISIN Mapping Process, AI automation micro-app: MF Portfolio Files Processor. Live app. Open source., Monthly MF portfolio files = hours wasted re-formatting. Here’s a tool that fixes it, Releasing MDRIFT - Mutual Fund Composition & Drift Analytics Tool"
},
{
"slug": "nifty50-30-day-forward-return-analysis-claude-in-excel",
"title": "NIFTY50 - 30 Day Forward Return Analysis Feb 2008 to 2026 - Claude in Excel with Python, Lambdas and Advanced Formulas",
"tags": [
"claude-in-excel",
"portfolio-analytics"
],
"content": "NIFTY50 - 30 Day Forward Return Analysis Feb 2008 to 2026 - Claude in Excel with Python, Lambdas and Advanced Formulas\n\nNIFTY50 Forward Returns Analysis\n\nFile shared + Power User Guide for Claude in Excel.\n\nAnalysis\n\nFor each Nifty50 trading day, compute 30 forward returns (Day 0 to Day 1, Day 0 to Day 2... Day 0 to Day 30). From those 30 returns - quintile cuts (P20, P40, P50, P60, P80) plus positive/negative day counts. And confidence intervals assuming normality for the time being. Then 30-day rolling diagnostics - max consecutive positive/negative streaks, max drawdown and runup %, days to max drawdown/runup, recovery days long and short.\n\nHow Claude handled it\n\nQuintile calculation in Python first, then validate with manual formula for a single day, then validate with LET + SEQUENCE, then setup a named LAMBDA function for reuse. Trade diagnostics using MAP, REDUCE and SCAN with LAMBDA - these walk through 30-day forward windows item by item. Had it setup documentation sheets ...plus a hyperlinked index sheet at start.\n\nWhole thing took just ~2.5 hours over two sessions. About an hour brainstorming and discussion, rest in processing, redoing, formatting. All done per instructions, but you need to be very clear ....and very granular sometimes in what you want it to do.\n\nStill WIP\n\n3 validations done for quintile cuts. Diagnostics need summarizing with distributions and confidence intervals, then full validation via Python recreation. After that - Gold, Oil & S&P 500 .. then add technicals ...then a regression...mostly quantile ...but yet to think through.\n\nFor traders\n\nThese are closing rates. Good for quick diagnostics only. Intraday prices needed for practical work. And for actual setups and backtesting, Claude Code / Scripts / Notebooks might be more efficient.\n\nBe aware:\nHeavy excel work is also token heavy\nPython code is reconstructed code, not the code snippets it actually ran.\n\nResources\nWorkbook: NIFTY50 Forward Return Analysis\nPower user guide - Claude in Excel: Power User Guide to Claude in Excel & PowerPoint - 26 Working Tips\nGet aligned price data: Portfolio Analysis Suite\n\n\nRelated\n\nTools: Quants Suite, QRep - Security Reports, Quants Agent\n\nExplore: Claude for Analytics hub, Quant tools hub, Python in Excel Practice Lab\n\nMore posts: Claude in Excel - Nifty50 Return Distribution Analysis (30 days forward) 2008 to 2026, U.S. Markets (S&P 500) vs India (Nifty 50) - is the returns profile reversing?, Power User Guide to Claude in Excel & PowerPoint - 26 Working Tips, Claude in Excel + MCP + xlwings Lite + Claude Code: Pull Data, Build a Model, Validate It."
},
{
"slug": "nse-sebi-surveillance-667-stocks-aug2026",
"title": "667 Stocks Are Under NSE and SEBI Surveillance. Almost All of Them Sit Outside the Nifty 500.",
"tags": [
"vigil"
],
"content": "667 Stocks Are Under NSE and SEBI Surveillance. Almost All of Them Sit Outside the Nifty 500.\n\nThere are 667 companies placed under surveillance by NSE & SEBI as of 7th Aug 2026. Is your stock among one of them?\n\nCheck it on TigZig's VIGIL app: vigil.tigzig.com. Updated daily.\n\nWhat is surveillance and how does it work: NSE Surveillance Measures - GSM, ASM, ESM Explained\n\nWhere the 667 actually sit\n\nSurveillance is largely a small and micro cap story.\n20 are in the Nifty 500. Of those, 2 are in the Nifty 100 and none in the Nifty 50.\n24 more are in the Microcap 250.\nThe remaining 623 are outside the Nifty Total Market, the 750 stock universe that is Nifty 500 plus Microcap 250.\nOf those 623, roughly 171 are on the SME platform and 41 are in the suspended or Z group. The rest is the long tail of the main board.\n\nPlus a company can carry several flags at once. They measure different things. ASM, GSM and ESM are price and volume surveillance. Separately there are fundamental ones. Inside the Nifty Total Market 750, 5 companies carry two flags and none carry three or more. Across all 667: 62 carry two, 27 carry three, 4 carry four.\n\nThe data is updated daily and you can download it manually from the app or pick it programmatically via API. Link below.\nvigil.tigzig.com/data-api"
},
{
"slug": "nyfed-household-debt-q2-2026",
"title": "NY Fed Q2 2026 Household Debt. The Headline Delinquency Flow Fell, but Auto and Mortgage Went the Other Way.",
"tags": [
"portfolio-quants",
"database-ai"
],
"content": "NY Fed Q2 2026 Household Debt. The Headline Delinquency Flow Fell, but Auto and Mortgage Went the Other Way.\n\nNY Fed Q2 2026 household debt was released last week. Overall 90+ DPD flow came down from 3.26% two quarters ago to 2.57%, however mortgage and auto went the other way, and against the pre-GFC years those two are the ones that look stretched.\n\nThe overall 90+ DPD flow, almost all of that is student loans unwinding after the reporting restart.\n\nMortgage is about 70% of everything households owe. Its 90+ DPD flow is 1.52%, back around the 2005 and 2006 range. The stock is 0.99% against 2.23% in Q3 2007, so on level mortgage is still a long way below the crisis, but that flow has crept up in six of the last seven quarters, and that's the pressure.\n\nAuto is in a red flag zone already. Auto 90+ DPD stock is 5.49%, the second highest reading in the whole series back to 2003, beaten only by last quarter. The flow is 3.00% against 2.34% in Q3 2007, and that has risen in each of the last two quarters.\n\nCards are above their 2007 level too, 6.97% flow against 6.52%, though cards have come down over the last two quarters.\n\nSeverely derogatory is 1.99% of all debt, a six-year high. Some of that is student defaults ageing in, and NY Fed does not publish that one by segment, so I cannot tell you how much.\n\nData is updated on the TREMOR app as well as the APIs, so you can use the interface to run your own cuts, or pick up the data via the API or download from the tool.\nTREMOR App: tigzig.com/tremor\nAPI Docs: tigzig.com/apis/tremor\n\nThe full comparison, today against Q3 2007\n\nThe quarter before the GFC delinquency cycle turned.\n\n90+ DPD stock, then to now\n\n| | Q3 2007 | Q2 2026 |\n|---|---|---|\n| Auto | 2.76% | 5.49% |\n| Card | 9.34% | 12.92% |\n| All debt | 3.06% | 3.31% |\n| Mortgage | 2.23% | 0.99% |\n\n90+ DPD flow, then to now\n\n| | Q3 2007 | Q2 2026 |\n|---|---|---|\n| Auto | 2.34% | 3.00% |\n| Card | 6.52% | 6.97% |\n| All debt | 3.24% | 2.57% |\n| Mortgage | 2.89% | 1.52% |\n\nSeverely derogatory 1.64% to 1.99%\n\nThe aggregate flow is actually below its 2007 level while the aggregate stock is above it, and mortgage is below on both. The elevated readings are coming from auto and cards.\n\nTool at TREMOR ➜ US Credit: tigzig.com/tremor\n\nWhat comes next, and why the three sources do not line up\n\nThe next two numbers on this are FDIC and NCUA, and both are due in the next few weeks.\n\nFDIC covers the insured banks, NCUA covers federally insured credit unions, and both of them split commercial from consumer.\n\nThe NY Fed panel works the other way round, it is built from credit files rather than lender filings, a 5% random sample of Equifax records, so it picks up consumer borrowing wherever it sits, including the part that has moved off bank balance sheets.\n\nThe definitions do not line up either. The NY Fed 90+ includes balances already charged off that are still on the credit report, where FDIC takes charge-offs out of delinquency and reports them separately as net charge-offs.\n\nBetween the three you get a reasonably full view of household and business credit. It is not complete, but it is closer than any one of them on its own.\n\nBoth FDIC and NCUA are still on Q1 2026 on the tool, so there is no Q2 read on the bank and credit union side yet. Both are on the same page and downloadable, and I will update them when they publish.\n\ntigzig.com\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QRep - Security Reports, DATS-4 Database AI Suite\n\nExplore: Tremor API docs, Analysis hub, API and MCP catalog\n\nMore posts: Point Your AI Agent at api.tigzig.com and It Works Out Which Dataset It Needs. 38 Million Rows Across India and the US, Free., Q1 2026 US Lending Data Live on Tigzig. NCUA + FDIC + NY Fed in One API. Two of Three Not on FRED. AI-Agent Ready., Auto Losses Past the GFC Peak. Cards Almost There. Credit Unions Already at GFC Loss Rates. NY Fed + NCUA Data Now on Tigzig., Technical Analysis on Any Ticker Without Visiting My Site. Just Ask Your AI Agent."
},
{
"slug": "oil-masking-effect-spot-futures-divergence-apr2026",
"title": "The Masking Effect in Oil Markets - The Price You See on Screens Is Hiding a $50 Supply Shock",
"tags": [
"portfolio-quants"
],
"content": "The Masking Effect in Oil Markets - The Price You See on Screens Is Hiding a $50 Supply Shock\n\nPublished: 2026-04-12\n\nLargest spot-futures divergence in oil market history. Brent futures at $94, physical crude at $144. Analysis of 9,864 trading days since 1987. 40 bids vs 4 offers in North Sea. Diesel near $200/bbl. US gasoline inventories at 16-year lows. Hormuz shut 40 days, Kuwait says 3-4 months to restore.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/oil-masking-effect-spot-futures-divergence-apr2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/oil-masking-effect-spot-futures-divergence-apr2026 with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/DATEDBRENT12APR2026.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nThe Oil Price You See Is WrongIt's heading for your paycheck and the global economy. The real price of crude crossed $144. Highest since 2008. Your screen shows $94. The gap is the largest in history\n\nAmar Harolikar\n\nDecision Sciences & Applied AI\n\n12th April 2026\n\ntremor.tigzig.com\n\n$94 - what your screen shows\n\n$144 - what refiners actually pay\n\nMar’26\n\nApr’26\n\nData & Tools\n\nData Sources\n\nBrent Spot (FOB) - EIA series (RBRTE) sourced via Refinitiv. Daily. Publication lag of 5-7 business days.\n\nBrent Futures - ICE front-month contract. Real-time via Yahoo Finance\n\nDated Brent (Physical) - S&P Global Platts daily assessment sourced from news reports from CNBC, Bloomberg, Reuters\n\nMethodology Note\n\nThe EIA Brent Spot series (RBRTE) tracks FOB NW Europe prices. This is not the same as Dated Brent (Platts), but both track closely. Platts Dated Brent hit $144 on April 8; the EIA series lags this.\n\nTool\n\nAll historical charts and data available on TREMOR (tremor.tigzig.com) with full download capability. News section aggregates headlines from 15+ financial sources.\n\nTREMOR tremor.tigzig.com\n\nTwo Oil Prices. Two Different Stories.\n\nBrent Futures (~$94-96) | What Yahoo Finance shows you\n\nA financial contract on ICE exchange for delivery next month or later. Millions of contracts traded daily. Heavily influenced by sentiment, ceasefire headlines, Trump statements. This is where hope lives.\n\nBrent Spot / Dated Brent (~$126-144) | The price of real oil\n\nThe price of actual crude you can load onto a ship right now. Published by S&P Global Platts. Set by real trades between refiners, traders, and producers in the North Sea. This is where reality lives.\n\nThe Largest Divergence in History\n\nIn 9,864 trading days since 1987, 94.7% had a spot-futures spread of $3 or less. Mean spread: $1.17. In March 2026, it exploded. Our data (EIA Brent Spot vs ICE futures) shows a peak of $22 on March 31 - already 2.3x the previous record. After April 2, per Bloomberg and S&P Global Platts, the gap widened further: Dated Brent hit $144 on April 8 while futures sat at $95 - a $49 gap.\n\n\"The financial market is almost masking the true tightness that everywhere else is showing up.\"\nAmrita Sen, Founder, Energy Aspects (CNBC, Apr 8)\n\n\"The paper traded markets are meeting physical reality.\"\nSultan al Jaber, CEO, ADNOC (Bloomberg via Yahoo Finance, Apr 11)\n\n$22 spread confirmed in our data (EIA Brent Spot vs ICE Futures, to Apr 2). $49 spread per S&P Global Platts Dated Brent vs ICE Futures on Apr 8, as reported by Bloomberg and CNBC.\n\nThe 40-Day Void\n\nHormuz has been effectively closed since early March. That's ~40 days of zero flow through a chokepoint that normally carries 20% of the world's seaborne crude.\n\nA supertanker from the Gulf to Rotterdam takes 20-25 days; to Japan or South Korea, 15-20. The last pre-war cargoes are arriving at refineries right now. Behind them - nothing.\n\nThis void doesn't disappear with a ceasefire. Even if Hormuz reopened today, the first new cargo wouldn't reach Europe until early May. Asia - mid-May. Gulf producers have shut down 13 million barrels/day of production because there's nowhere to send it. That's not capacity that switches back on at a press conference.\n\n\"The paper traded markets are meeting physical reality, and the 40-day gap in global energy flows is truly exposed.\"\nSultan al Jaber, CEO, ADNOC (Bloomberg via Yahoo Finance, Apr 11)\n\n\"Perceived geopolitical risk can ease faster than operational risk.\"\nJaniv Shah, VP Oil Markets, Rystad Energy (CNBC, Apr 10)\n\nThe Global Scramble for Barrels\n\nIn the North Sea this week, traders submitted 40 bids for cargoes. Only 4 were met. Asian refiners told Bloomberg they were \"no longer focused on price\" - just securing supply. The scramble is global:\n\nJapan - buying US crude on smaller ships\n\nChina - buying spree pushed Vancouver oil shipments to a record high this month\n\nIndia - doubling Venezuelan purchases;\n\nUS - exporting at record levels; WTI premium quadrupled to ~$4/bbl\n\nNigeria - crude premiums at $25/bbl above benchmark vs <$3 before the war\n\nNorth Sea - Trafigura and Gunvor bidding $22/bbl above Dated Brent for late April delivery\n\n\"There is simply a shortage of crude. Physical Brent is a mess. At this rate even European refiners will have to lower utilization, perhaps as early as next month.“ - Neil Crosby, Head of Research, Sparta Commodities (Bloomberg via Yahoo Finance, Apr 11)\n\nThe US Is Exporting Its Own Cushion\n\nRight now the US looks like the winner. Record crude exports. Tankers lining up at Houston. But every barrel exported is a barrel US refineries don't process - and the math is turning.\n\nUS gasoline inventories have shrunk to the smallest in almost 16 years per the EIA. Jet fuel and diesel have soared to record highs above $200/bbl. The crude leaving US ports is being replaced by nothing - because nothing is coming in from the Gulf.\n\nThe Midland WTI premium at Houston has quadrupled - not because US crude is suddenly more valuable, but because Houston is one of the few places on earth you can actually load a tanker right now. That's a transit premium, not a quality premium. The US is becoming the world's gas station of last resort - and gas stations run dry when everyone shows up at once.\n\n\"If futures don't catch up to the physical realities, US exports could easily remain elevated to the point where there isn't enough crude left for US refineries.“ - Amrita Sen, Co-founder, Energy Aspects (Bloomberg via Yahoo Finance, Apr 11)\n\nThis is how a supply shock in crude transmits into the real economy - not through the futures price on Yahoo Finance, but through the physical price that refiners actually pay.\n\nRefiners who can't find crude or can't afford it are cutting production. European refiners are expected to lower utilization as early as May. Asian refiners have already started. When refineries cut runs, less diesel, jet fuel, and gasoline gets produced. The crude shortage becomes your fuel shortage. Diesel and jet fuel have already hit record levels above $200/bbl.\n\nA financing problem makes it worse. A refiner buys physical crude at $130, hedges with futures at $95. On paper the margins work. In reality, the $35 gap is a cash outflow they must finance upfront. A Shell or ExxonMobil can handle this. Smaller refiners don't have the balance sheet. So they step back - and your fuel gets more expensive.\n\n\"On paper the margins are fantastic, but the real cashflows of buying a cargo and deciding to refine it can be quite different.“ - Roberto Ulivieri, Consultant, Midhurst Downstream (Bloomberg via Yahoo Finance, Apr 11)\n\nFrom Blocked Strait to $200 Diesel\n\nEven If Hormuz Opens Tomorrow\n\nThe ceasefire is two weeks old and fragile. But assume the best case - full reopening, normal traffic resumes. Even then:\n\nProduction restart: 3-4 months. Gulf producers shut 13M bpd. Kuwait Petroleum CEO: \"bulk within weeks, full production within three or four months.\"\n\nTanker redirect: until June. Most of the world's tanker fleet is currently pointed toward the US. Redirecting back to the Gulf takes months.\n\nInfrastructure damage: real. Saudi attacks reduced capacity by 600K bpd, cut East-West Pipeline throughput by another 700K bpd - the pipeline built specifically to bypass Hormuz.\n\nFull restoration: ~5 months. Hundreds of millions of barrels taken off the market. Kpler's Amena Bakr estimates five months to fully restore capacity.\n\n\"The bulk will come within a few weeks, and then the full production will come within three or four months.“ - Sheikh Nawaf al-Sabah, CEO, Kuwait Petroleum (CNBC, Apr 8)\n\nThe best-case scenario is still months of pain. The worst case is significantly worse.\n\nOil Markets : Hope vs. Reality\n\nFutures respond to tweets, ceasefires, and diplomatic optimism. Dated Brent responds to how many barrels are actually on ships right now.\n\nTraditional crude relationships have broken down entirely. Patterns that held for decades flipped in weeks:\n\nBrent vs WTI: Brent normally trades $3-5 above WTI. During the crisis, WTI traded at a $10+ premium over Brent - a complete reversal.\n\nRussian Urals: Traded at steep discounts to Brent since 2022. In recent weeks, Urals reached $30 above Brent.\n\nSaudi Arab Light: Premium over Oman/Dubai raised to $19.50. This premium had never before exceeded $10.\n\n\"This speaks to unprecedented stress and uncertainty in the oil market.\"\nPavel Molchanov, Senior Analyst, Raymond James (CNBC, Apr 10)\n\nHow a Blocked Strait Hits Economies\n\nHormuz doesn't just carry crude. It carries 20% of global LNG, nearly half the world's urea, 30% of ammonia (fertilizer raw materials), a third of global helium, and 20% of raw aluminum exports. The transmission chain is already in motion:\n\nFuel: US gasoline at $4.10/gal national average. California above $5. Diesel and jet fuel near $200/bbl.\n\nFertilizer: Urea prices up 50% since the war started. Planting season is now. Farmers pay more. Food costs more later.\n\nFreight: Ships rerouting around the Cape of Good Hope - adding weeks to transit and billions to global trade costs.\n\nInflation: Oil shock feeds directly into CPI. Fed rate cuts become harder to justify.\n\nGrowth: EU countries cutting 2026-27 GDP projections. Higher energy costs + tighter monetary policy = the classic stagflation setup.\n\nThis is what the futures market at $95 is not pricing. A barrel of crude at $95 tells one story. A barrel of diesel at $200, fertilizer up 50%, and a container ship adding 3 weeks to its voyage tell a completely different one.\n\nHow a Blocked Strait Hits Economies\n\nThe closure doesn't need to last forever to cause lasting damage. It just needs to last long enough for these transmission chains to embed in prices. That process has already started.\n\nThree Scenarios – One Thing in Common\n\nReferences\n\nTREMOR | Macro Signals, Data & Charts\n\nBloomberg, A Panicked Race for Barrels Is Gripping the Global Oil Market, April 11, 2026\n\nCNBC, Brent oil spot price above $120 in sign that Iran ceasefire can't solve deep disruption, April 8, 2026\n\nCNBC, What this real-world oil price says about the level of stress in the energy market, April 10, 2026\n\nOilPrice.com, North Sea Crude Soars to Record High as Hormuz Shock Rips Through Spot Markets, April 11, 2026\n\nS&P Global, Platts Dated Brent Price Assessment\n\nEIA, Spot Prices for Crude Oil and Petroleum Products\n\nRun Your Own Analysis\n\nTREMOR : Macro Early Warnings Signalstremor.tigzig.com\n\nCharts - Overlays, MoM, YoY\n\nAI Support\n\nData Download\n\nNews\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: Quant Apps hub, Analysis archive, APIs and MCP\n\nMore posts: Three Red Flags Hit Together. 30Y at 5.14%, PPI at 5.99%, Brent Above $100. Setup Worse Than 2000 or 2007., Are We Headed for Stagflation-Lite, Which Chart Is the Odd One Out? Three Macro Charts Tell the Same Story. One Doesn't., TREMOR - Macro Early Warning Dashboard - 50+ Indicators, Updated Daily"
},
{
"slug": "open-so",
"title": "Live Portfolio Analytics - Powered by MCP Servers - Open Source",
"tags": [
"portfolio-analytics",
"mcp"
],
"content": "Live Portfolio Analytics - Powered by MCP Servers - Open Source\n\nAnalyze any stock, crypto, metal, or oil symbol vs benchmark using 70+ KPIs & 15+ Charts, AI-powered technicals, and clean PDF + web reports.\n\nLive across 6 interfaces: web apps, ChatGPT, chat agents, Excel (xlwings Lite), and forms.\n\nEach interface serves a different use case - from rapid-deploy to full-featured agents. Modular architecture makes it easy to plug components into any flow - agentic or not.\n\nBuilt the backend MCP servers, agent flows, and user interfaces as reusable modular components - easy to plug into different use cases and mix across stacks.\n\nPerformance stats powered by Python QuantStats package (by Ran Aroussi, creator of yfinance). Technical chart analysis with Python Finta and Gemini Vision. MCP servers built with Tadata FastAPI-MCP package, web apps with React and NextJS, and MCP enabled agents on n8n & Flowise.\n\nFully modular and live - clone it, remix it, set up your own stack\n\nMCP Servers are public:\nQuantStats MCP\nYFinance MCP\nTechnical Analysis MCP\n\nFull build breakdown below\nWHAT IT DOES - Live Portfolio Analytics Stack\n\nThis is a working analytics stack delivering live performance reports, AI-driven technicals, and financial data pulls - powered by MCP, FastAPI, and modular agents.\nQuantStats Performance vs. Benchmark - For any Yahoo Finance symbol - stocks, crypto, oil, metals. Over 70 metrics, 15+ charts: drawdowns, Sharpe/Sortino ratios, return distributions, correlations - delivered in clean HTML\nAI Technical Analysis - Technical analysis across two timeframes - Across daily and weekly timeframes. Chart analysis via Gemini Vision API, delivered as PDF and web reports with structured tables.\nFinance Data Pull: Extract prices, profiles, and full financials from Yahoo Finance: 150+ fields of profile info, P&L, balance sheet, cash flow. Excel integration via xlwings Lite\nHOW TO USE - 6 Live Interfaces\n\nThe system runs across 6 interfaces - each tailored for different use cases\nAgentic - Custom UI (NextJS + n8n)\nAgentic - ChatGPT (Custom GPT)\nAgentic - Advanced (React-based with full analysis tools)\nAgentic - Rapid Deploy (Flowise Native)\nExcel - xlwings Lite\nForm UI - HTML-JS-Jinja2\n\nAsk the agent for guidance or start with a prebuilt prompt.\n\nAll support the same analytics setup - just different frontends and feature layers. See next section on interface strategy.\nINTERFACE STRATEGY - Why 6, and Why Modular\n\nThe setup is designed around modular Gen AI-powered components - backend, agent layer, and UI - each one reusable and configurable depending on the use case. Once core processing is in place, it's easy to plug into different interfaces without rebuilding the logic.\n\nThe six interfaces aren't just demos - they show real deployment options. From lightweight forms to full-stack apps and AI agents. The options support a wide range of use cases - Section 6 on User Interfaces goes into details on when to use which and the trade-offs involved\n\nTo support this, I built three MCP-FastAPI servers and one standalone FastAPI server. These connect to agents running on n8n and Flowise, and frontends on React and Next.js. All components are connected via standard APIs, making them portable across tools - including third-party platforms.\n\nIn practice, modularity isn't always necessary. I sometimes deliver integrated solutions where the UI, logic, and agent live in the same build - faster for simpler use cases. But where reusability or scale is a factor, modular saves time, simplifies updates, and isolates risk.\n\nThis isn't about MCP or agents - it's about building practical, reusable analytics solutions that can plug into any interface or automation flow.\n\nAll live. Test it, clone it, build your own.\nARCHITECTURE - Modular, Component based, Reusable\n\nThe full stack is built around reusable components - each layer (frontend, agents, backend) is designed to plug into others with minimal setup. Here's how the architecture breaks down across interfaces and agents.\n\n4.1. Modular - Component based: Why and When?\n\nThere are cases where a non-modular setup makes more sense. For example, in one client project I built a lightweight HTML-JS tool for Excel file import and manipulation, bundled with a browser-based SQLite agent and a simple NL-to-SQL chat interface - all integrated in a single app. In that setup, modularity would've just added unnecessary complexity.\n\nBut when I see components - UI, backend logic, agents - that can be reused, I default to separating them out. In another case, I built a custom UI connected to a Flowise agent and backend SQL service. Later, the client needed a second agent setup pointing to a different database. All I had to do was update the agent config and env variable - no UI rebuild, no backend changes.\n\nModular setups also help with debugging, iteration, and access control. I can isolate issues, restrict backend exposure, and upgrade parts independently.\n\nI started building my technical analysis stack directly inside Excel with xlwings Lite. As it evolved, I split core processing into an MCP-FastAPI server - now the same logic runs across all UIs: web, forms, agents, GPT, Excel.\n\nNone of this is new - tech has done it for decades. Just sharing how modularity speeds up my own analytics builds, and when I choose to keep it simple.\n\n4.2. Frontend options\n\nCovered in detail in a separate section with notes on when to use what, and trade-offs based on my experience.\n\nUI options include: Next.js, React, ChatGPT, form-based UI, Flowise native UI, and Excel\n\nReact and NextJS UIs are set up as reusable modules - they can connect to agent flows on Flowise, n8n, or any API-accessible setup. Just update the API endpoint in the env variable, match input/output formats, and it's live.\n\nIn the current setup, the n8n agent connects to the NextJS UI, and the Flowise agent connects to the React app. Both agents are MCP enabled and are in turn connected to MCP Servers.\n\n4.3. Agent setups\n\nThe agent layer acts as the glue between UIs and backend MCP servers - handling workflows, orchestration, and logic routing. Here's how the n8n and Flowise setups are configured.\n\nn8n and Flowise AI support webhooks and API endpoints - easy to plug into any interface and is great for keeping the Agent layer separate.\n\nBoth include MCP Clients with SSE support for remote MCP Servers. Just drop in the MCP Server URL and you're set. In the current setup, agents connect to multiple MCP Servers and database tools (Flowise).\n\nBoth are production-grade tools built for complex agentic and non-agentic workflows (n8n). Flowise supports Sequential Agents with LangGraph - enabling advanced orchestration with routing and step-wise execution. n8n is superb with its Agent node, wide platform integration, HTTP node for API calls, and a solid set of routing and processing nodes.\n\n4.4. Core engine\n\nThis is the processing brain of the system - everything from calculations to report formatting runs through this Python-based backend, wrapped in modular MCP-FastAPI services.\nQuantStats by Ran Aroussi (creator of yfinance)\nyfinance for market data\nFinta for technical indicators\nMatplotlib for charts\nGemini Vision API for visual chart analysis\nReportLab for PDF formatting\n\n4.5. Backend\n\nThree integrated MCP + FastAPI servers and one standalone FastAPI server (details in next section). All Python logic is wrapped in FastAPI and mounted on an MCP Server in a single deployment - which also serves the form UI.\n\nI keep all reusable logic on FastAPI - easy to automate and connect across UIs, from ChatGPT to Excel to any custom UI. Tadata's FastAPI-MCP package makes it simple to mount MCP on any existing FastAPI setup.\n\nConnections:\nn8n and Flowise agents connect to the MCP server via their native MCP Client nodes\nReact and NextJS UIs connect to agents via API\nChatGPT connected to FastAPI endpoints on the same integrated MCP-FastAPI server via Custom Actions OpenAPI schema\nForm UI connected to FastAPI endpoints on the same MCP-FastAPI server\nExcel connects to FastAPI endpoints through xlwings Lite\nMCP SERVERS\n\nThe backend is split into four focused processing services - each one handles a specific piece of the analytics workflow, from financial data to report generation. All are exposed via MCP or API, built for reuse and quick integration.\n\nThere are three custom MCP Servers and one standalone FastAPI Server. The MCP servers are public - just plug and play. Add the URL to any SSE-enabled MCP client. Both n8n and Flowise have native nodes for this.\n\nThe servers are integrated MCP-FastAPI servers. I used Tadata's FastAPI-MCP package to mount MCP on top of FastAPI - just a few lines of codes - brilliant package. A single deployment runs MCP + FastAPI + Form UI (HTML-JS-Jinja2). Works cleanly with both agentic and non-agentic setups.\n\nAll the user interfaces connect to these integrated MCP-FastAPI servers - explained in the User Interfaces section\n\n5.1. QuantStats MCP Server\n\nThis is an MCP-FastAPI wrapper over the QuantStats package - connects to any frontend via MCP or API. Takes two Yahoo Finance symbols (one for performance, one for benchmark) plus a time range. Returns a formatted HTML report using the QuantStats package. Tables and charts are auto-generated directly by the package.\n\nThe MCP enabled agents connect to MCP Server and rest of UIs to the FastAPI endpoints.\n\n5.2. Technical Analysis MCP Server\n\nThis is a processing server that runs a multi-step workflow to generate technical analysis reports in PDF and web format. Takes a Yahoo Finance symbol and time range, returns AI-generated reports.\n\nWorkflow steps:\nConnects to Yahoo Finance MCP-FastAPI server to pull price data\nConverts daily prices to weekly dataframe\nCalculates technical indicators using Finta\nConnects to Gemini Vision API to get chart and technical analysis\nConnects to ReportLab FastAPI server for generating the final PDF and Web reports.\n\nThe MCP enabled agents connect to MCP Server and rest of UIs to the FastAPI endpoints.\n\n5.3 Yahoo Finance MCP Server\n\nThis is an MCP-FastAPI wrapper over the yfinance package. Connects to any frontend via MCP/ API. Takes a Yahoo Finance symbol and returns:\nPrice data (JSON) - requires date range\nCompany profile with 150+ fields\nFull financials: P&L, balance sheet, cash flow, and quarterly breakdowns\n\nThe MCP enabled agents connect to MCP Server and rest of UIs to the FastAPI endpoints.\n\n5.4 ReportLab Markdown to PDF-HTML FastAPI Server\n\nThis is a FastAPI processing server for generating custom formatted reports. PDF and HTML outputs are customized for this use case but can be adapted for others. ReportLab offers deep customization for PDFs, easily replicated in web reports using standard HTML-JS.\n\nWorkflow:\nConvert Markdown to HTML (using markdown)\nParse HTML with BeautifulSoup for structure\nUse ReportLab to build styled PDF\nStyle HTML output to match PDF\nReference charts from static folder\nAuto-clean old files (older than 24 hrs) using FastAPI startup events + Starlette background tasks\n\nI've set up a separate endpoint for technical analysis, with custom formatting for both PDF and HTML outputs. The same FastAPI server also includes a generic endpoint that takes Markdown content and returns a PDF with simpler formatting. It's deployed as an HTML-JS-Tailwind form UI that calls FastAPI endpoints - all served from a unified FastAPI server using Jinja2 templates.\nUSER INTERFACES\n\nAll processing and agent logic connects to live, working interfaces. Each UI is connected to the same backend and agent layer - just optimized for different workflows, tools, or user preferences. All of these are live, working apps - each built on top of the same backend stack. The UI is just the entry point. Some are lightweight and fast to deploy, others offer more control or customization.\n\n6.1. Custom GPT\n\nCustom GPT is usually my first choice when a UI is needed. You get an out-of-the-box interface, embedded agent, and Python code execution. Just supply a JSON OpenAPI schema to connect to any backend. Faster and cleaner than building even basic HTML-JS forms.\n\nLimitations: no custom UI, and feature set is narrower than a React or NextJS build. You'll need a Plus account ($20/month) to create a Custom GPT, though free users can still access it with rate limits.\n\nLive setup connects to QuantStats and Technical Analysis MCP-FastAPI servers via FastAPI endpoints.\n\n6.2. Flowise native UI\n\nFlowise native UI is a solid option when GPT isn't feasible. No UI build required. You get a ready-to-use chat interface that supports complex agent flows using LangGraph Sequential Agents, custom tools, and APIs. n8n too provides a similar native UI.\n\n6.3. Excel integration - xlwings Lite\n\nxlwings Lite (by Felix Zumstein creator of xlwings) is a lightweight Excel add-in that runs full Python workflows inside Excel - no local Python install needed. It comes with a built-in editor, console, environment variables, and deep Excel integration.\n\nOne of the big benefits is that it allows easy connectivity to any CORS-enabled backend API service. This fits well with my setup - since I use FastAPI servers extensively - and also makes it easy to connect to LLM/AI endpoints and third-party server providers.\n\n6.4. Simple Forms\n\nIn many cases, a simple form works best. I typically use Jinja2 templates (HTML-JS-Tailwind) for tight FastAPI integration. Sometimes Flask. Works well for fairly complex UIs too, with full JavaScript access and server-side rendering - env vars and routes stay hidden.\n\n6.5. Going full stack? React / NextJS\n\nFor complex apps and a polished UI, I use React or NextJS. The REX-3 Co-Analyst app has the QuantStats Agent on a separate tab, connected to the same Flowise agent from earlier - just wrapped inside a React interface.\n\n6.6. NextJS\n\nThe biggest benefit of NextJS: env vars and API routes stay private, and it supports server-side rendering. That's a major security benefit. The NextJS portfolio analysis agent uses a leaner UI, but it's still set up in a modular way - can be connected to any API-enabled agent backend by just changing the environment variable.\n\nNote: Vercel serverless functions time out at 60 seconds (300s on pro plan), so longer API calls need workarounds.\nDEPLOYMENTS\n\nNextJS and React apps are deployed on Vercel. All MCP + FastAPI servers, along with n8n and Flowise, run on a Dockerized setup via Coolify, deployed on a Hetzner VPS.\nAI CODER: CURSOR\n\nI use Cursor as my AI coding assistant across all builds - including this one. Every part of this stack - from UI's to FastAPI servers - was written and iterated using Cursor.\n\nFine print: This is not investment research or financial advice. It's a live working example showing how to stitch together AI, analytics, and infra into real outputs. The logic and analysis structure is based on a general-use setup - fully modifiable to fit your own requirements. Source code, backend, and app stack are open and adaptable. AI and humans can both make mistakes - always validate results.\n\n\nRelated\n\nTools: Quants Agent, QRep - Security Reports, Quants Suite\n\nExplore: Quant Apps hub, MCP hub, Practice Lab\n\nMore posts: Three Tools I Built for Market Analysis - QSUITE, TREMOR, QREP, Build AI Workflows with MCP Servers + n8n, ChatGPT Connected to integrated FastAPI-MCP Servers.. Technical Analysis (TA) report. From stocks to crypto., Quants Agent: Now with LLM Choices for Technical Analysis Reports, QRep - Powered by QuantStats. Live Now."
},
{
"slug": "open-source-asset-comparison-tool-compare-stocks-indices-crypto-commodities-in-one-dashboard",
"title": "Open Source Asset Comparison Tool: Compare Stocks, Indices, Crypto & Commodities in One Dashboard",
"tags": [
"portfolio-analytics"
],
"content": "Open Source Asset Comparison Tool: Compare Stocks, Indices, Crypto & Commodities in One Dashboard\n\nEver tried comparing Apple, TCS, Nifty, Gold & Oil in one report - without juggling multiple platforms or drowning in paid apps?\n\nI use Trading View regularly. Used Amibroker back in my full-time trading days. Koyfin, Portfolio Visualizer, Portfolio Metrics - solid tools.\n\nBut here's my problem: I don't want model portfolios.\n\nI want to compare across markets and assets classes - Nifty vs Gold vs Microsoft vs Brent. Fast.\n\nWhat I wanted\nDaily returns chart - all securities in one chart\nDrawdowns, CAGR, Sharpe, Sortino - single visual\nDate-aligned CSV downloads\nTechnical analysis report\nRun combos of assets and timeframes\nAll in 30 seconds, not hours. In same tab.\n\nTry doing that on free tools - you'll hit limits, create 'portfolios' for single stocks, run into missing securities or juggle multiple tabs.\n\nSo I built TIGZIG Quants\n\nMy workflow now:\nQuick comparison on TIGZIG Quants\nDeeper TA -> Trading View\nDeeper Research -> Platforms and offline work\n\nIs this the ultimate solution? Absolutely not!\n\nBut it solves my problem: fast, no-nonsense comparisons between unrelated assets.\n\nQuick disclosure: I'm a Chartered Accountant and data scientist, not a portfolio manager. I built this for my personal research, and while it's a powerful tool, it isn't investment advice. Do your own due diligence always\n\nThis is build-as-you-go. Right now, CAGR is point-to-point (not a fan - too sensitive to endpoints). Rolling metrics & ML forecasts coming soon.\n\nReady to try it?\nAgent - Smart assistant\nSuite - Faster, no-frills version (needs exact Yahoo symbols)\n\nWant to build your own version? Hit the Doc button in the app for source codes and build guides + methodology, validations and recon report. Built with MCP servers, n8n, and React.\n\n\nRelated\n\nTools: Quants Agent, Quants Suite, QRep - Security Reports\n\nExplore: Quant Apps hub, MCP servers directory, Guides and analyses\n\nMore posts: Three Tools I Built for Market Analysis - QSUITE, TREMOR, QREP, Quants Suite. 5 Reports. Performance, Risk & Technical Analytics., Build AI Workflows with MCP Servers + n8n"
},
{
"slug": "oracle-always-free-arm-vps-retry-script",
"title": "How to Get Oracle's Always Free ARM VPS (24GB RAM, 4 CPUs). Auto-Retry Script to Beat 'Out of Capacity'.",
"tags": [
"infrastructure"
],
"content": "How to Get Oracle's Always Free ARM VPS (24GB RAM, 4 CPUs). Auto-Retry Script to Beat 'Out of Capacity'.\n\nOracle ARM VPS Monitor\n\nOracle Cloud's Always Free tier gives you one ARM VPS, the Ampere A1 (VM.Standard.A1.Flex) shape, with 24GB RAM, 4 CPUs and 200GB storage. Forever free. Like an EC2 or Azure VM. The only cloud provider that gives this away always free.\n\nBanged my head for over a year on the web console. The instance is almost always out of capacity across every availability domain.\n\nFinally asked my AI Coder (Claude Code now) to research how people actually get these instances. Answer: automated retry scripts running 24/7, cycling through availability zones. Takes 1-3 months typically.\n\nHad it build a retry script with monitoring dashboard, deployed on my server. The script will keep retrying until a VM is created. Results are saved to json on server on success. Checks all 3 availability domains each round. Ran 30+ days, over 100K API calls. Finally got the instance. Felt like I won a lottery.\n\nOracle reclaims inactive free instances - so put it to work immediately. Had Claude Code secure the server, install Coolify, deploy my DuckDB dashboards as alternate backend (IMDb data, 230M rows, 16GB database, FastAPI layer) - same day.\n\nResources below - VPS creation tool, infra guides, live dashboard.\n\nBonus: Oracle also gives 2 AMD Micro VMs (E2.1.Micro) - 1 CPU, 1 GB RAM each, x86 (not ARM), always available. No lottery. Great for many tasks. Each uses ~50 GB of your 200 GB quota.\n\nResources\n\nOracle ARM VPS Creator: Real-time Web UI with Live Monitoring\nhttps://github.com/amararun/oci-arm-vm-monitor\nFastAPI + HTML interface with Basic Auth. All OCI credentials stay server-side.\n\n2026 Infra Guide for Tool Builders\nPart 1: AI Coders\nPart 2: Deployment & Hosting\nPart 3: Security\n\nLive Dashboard (DuckDB, IMDb data, 230M rows, 16GB, Open Source)\nhttps://www.tigzig.com/movie-explorer\nToggle between Hetzner/Oracle servers.\n\n\nRelated\n\nTools: Movie Explorer, TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Browse all apps, Database AI apps, Guides hub\n\nMore posts: You are paying ~$3-7 per deployment for your AI Apps. How do you do it in <$10 per month?, Full backend Infrastructure for DuckDB production and analytics system, CinePro - 230M Rows, 16GB Database, Instant Queries with DuckDB, 2026 Infra Guide for AI Tool Builders - Part 2: Deployment & Hosting, Server Meltdown: How Bots Crashed My AI Tools and What I Did About It"
},
{
"slug": "perimeter-security-cloudflare-free-plan",
"title": "2026 Infra Guide for AI Tool Builders - Part 5: Perimeter Security on Cloudflare (Free Plan)",
"tags": [
"cloudflare",
"security",
"infrastructure"
],
"content": "Infra Guide for AI Tool Builders - Part 5: Set Up Perimeter Security (Edge Defense) for Your Apps on Cloudflare's Free Plan\n\nEdge Security Cloudflare Free Plan\n\nPrevious posts covered frontend, backend, and database security. This one is about what happens before traffic even reaches your server. Edge defense. Perimeter security. The stuff that sits between the internet and your application.\n\nIf you come from an analyst or data science background like me — building tools and small apps for businesses — you may not know about Cloudflare capabilities. I didn't. Even on the free plan, the amount of security and non security functionalities it offers is huge. And it costs nothing.\n\nHere's what I'm running across 60+ subdomains and 40+ deployed apps. All on the free tier.\n\nJust note that, with Cloudflare tokens, your AI Coder (Claude Code, Cursor etc) would be able to do any / all of this for you.\n\nOrange Cloud Everything\n\nFirst thing. When you add a domain to Cloudflare, every DNS record gets a toggle — orange cloud (proxied) or grey cloud (DNS-only). Orange cloud means traffic flows through Cloudflare. Grey cloud means traffic goes directly to your server, bypassing everything. Every domain that can be orange-clouded should be. The exceptions are third-party CNAMEs — Auth0, Clerk, AWS RDS — those need grey cloud because the third-party service requires direct resolution.\n\nOne issue I faced here - It has a 100 seconds time out. For my longer running backends - they are temporarily on grey cloud as I migrate to a polling system .. i.e my backend will send a job ID and I'll poll the completion endpoint from my frontend with that job ID.\n\nImportant thing is - know exactly which domains are grey and why. I maintain an audit list. Everything else — orange.\n\nJS Challenge on Frontend Domains\n\nCloudflare gives you 5 free custom WAF (Web Application Firewall) rules. One of mine applies a JavaScript challenge to most of my frontend tools. What this does — when a request hits your site, Cloudflare injects a lightweight JS snippet that the browser solves transparently. Users barely even notice. But automated scanners, bots, and curl scripts can't solve it. They get blocked before touching your app.\n\nBut here's an issue I faced- tigzig.com is an AI Agent first site with the all domains assets available as text and markdown files and indexed via llms.txt. But with JS Challenge I suddenly saw a drop in AI Agent hits, even though the llms.txt and other assets were excluded from the challenge and were fully accessible. My hypothesis - AI Agents skip the site without bothering to check for llms.txt if they hit a JS challenge on main domain. And as soon as I removed the JS Challenge, they came happily back.\n\nBrowser Integrity and Security Level\n\nFree settings. Browser Integrity Check catches requests with fake or missing User-Agent headers. Security Level set to High makes Cloudflare challenge IPs with bad reputation scores. Two toggles. Takes 30 seconds.\n\nThe Rate Limit Problem\n\nCloudflare's free WAF gives you one rate limit rule. One threshold across your entire zone. But a frontend React app could send anywhere from 20-40 requests on page load — JS bundles, CSS, images, API calls. For backend API, it's much lesser as it receives only the calls from frontend. So you need different rate limits for frontend and backends.\n\nThis is where Cloudflare Workers come in. Free plan gives you 100,000 invocations per day. You upload a small script that runs at the edge — before traffic reaches your server. My Worker checks the incoming domain against a config map and applies different rate limits:\nFrontends: higher limit - handles page load plus a few refreshes\nBackends: lower limit - pure API, no page loads\nMCP servers: medium limit - as AI agents can shoot off queries fast\nCatch-all: medium limit for any that might remain unconfigured\n\nOne Worker. One deployment. Different limits per domain and even per path. When the limit is hit, Cloudflare returns a 429 at the edge. The request never reaches your server. Your AI coder can write and deploy this — it's a single JavaScript file with a configuration map.\n\nAnd you can monitor your 429 rejects on Cloudflare dashboard and calibrate limits after having your AI Coder investigate their payload. I monitor this using a custom dashboard.\n\nIs the 100K invocations enough?\n\nIt looks like a big number but actually not so. Try this - next time you deploy a new app - check the number of requests at Cloudflare. Within first day you'll see thousands of hits on Cloudflare even without a single page view on your web analytics dashboard. Why? All automated bot scanners doing vulnerability scans. For a large public facing app, that's a drop in the ocean. For internal tools for small businesses, that might work....it all depends on what all you got in that particular zone. So yes, monitoring your free quota usage is important.\n\nWhat if you run out of free quota? Then there is option of Workers paid plan of $5 p.m which covers 100 million requests/month + $0.30 per additional million.\n\nFree tier is also limited to 10ms CPU time (not wall clock time) per request. For rate limit checks - domain lookup, config match, counter comparison - that's more than enough.\n\nYou need to monitor usage. A single bot storm could eat through that quota fast.\n\nZone-Level IP Blocking\n\nWhen you spot a malicious IP in your logs — vulnerability scanners, credential stuffers, path traversal attempts — you can block the entire IP range at the Cloudflare zone level (either a single IP or an entire subnet say, 256 addresses owned by the same provider). One API call. Applies across all subdomains instantly. I have blocked a few ranges from a bulletproof hosting provider (hosting provider known for harboring malicious traffic) after they hit my MCP server with .env hunting and malicious file upload attempts. Every request from those IP ranges now gets dropped at the edge. Zero impact on my servers. This is only as an exception. You can't really keep blocking IPs as they get rotated. The scalable way is rate limits, JS challenges etc as well as hardened & non-leaky backends. But good to know in case you do need to do that.\n\n\n
\n\nVercel .vercel.app Bypass\n\nIf you're on Vercel, every app gets a public something.vercel.app URL alongside your custom domain. This URL bypasses Cloudflare entirely — all your edge rules become useless. Fix: enable Deployment Protection in Vercel settings. Any access to the .vercel.app URL goes to Vercel's sign-in page. One toggle. Matter over.\n\nWhat Comes After the Edge\n\nOnce traffic passes through Cloudflare, it hits your server. That's where the other 80+ items in my security checklist kick in — frontend hardening, backend rate limits, SQL validation, API key auth, database read-only roles, and so on. The edge is the first gate.\n\nMy full security checklist — 95 items across React, FastAPI, Postgres, DuckDB, Cloudflare, MCP servers, VPS, and auth — is at tigzig.com/security. Each of these detected in my own apps and in process of fixing. There's a copy-to-markdown button. Paste it to your AI coder and have it walk through your app. Or use it as a starting point and customize.\n\nTotal cost of everything above: $0. All on Cloudflare's free tier. The only cost is the time to set it up - and your AI coder handles nearly all of that.\n\nSecurity Checklist for Web Apps — Check list here: 95 items, all practical stuff tigzig.com/security\n\nPost here: Security Checklist for Web Apps\n\nMCP Server security with OAuth & hardened backends: Talk to Your Database from Excel - Part 2\n\nInfra Guide Series\n\nThe full \"Infra Guide for AI Tool Builders\" series:\nPart 1: AI Coder - Claude Code across the full build-to-deploy chain\nPart 2: Deployment & Hosting - Hetzner, Coolify, Vercel, Cloudflare\nPart 3: The 18 Common Security Mistakes (and Fixes) - production security hardening\nPart 4: Free CORS Proxy - Cloudflare Worker Code - copy-paste Worker to bypass CORS\nPart 5: Perimeter Security on Cloudflare (Free Plan) - edge defense, WAF, rate limits, honeypots\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Security checklist, MCP + APIs hub, Browse all apps\n\nMore posts: Per-IP rate limit on Cloudflare free plan - the gotchas, 2026 Infra Guide for AI Tool Builders - Part 3: The 18 Common Security Mistakes and How to Fix Them, My Public MCP Server Got Hammered - Security Lessons from a Bot Attack, Security Checklist for Web Apps - 71 Items, Are You Rate Limiting the Wrong IPs? A SlowAPI Story."
},
{
"slug": "portfolio-agent-mutual-fund-nav-aug2026",
"title": "My Portfolio Agent Could Not Look Up an Indian Mutual Fund NAV. Now It Can.",
"tags": [
"portfolio-quants",
"mutual-funds"
],
"content": "My Portfolio Agent Could Not Look Up an Indian Mutual Fund NAV. Now It Can.\n\nSomeone asked my portfolio agent to look up a mutual fund scheme and give them its NAV, and the agent had to tell them it could not. Now it can.\n\nWhat happened was that the agent had every other kind of market data wired into it and nothing at all for Indian mutual funds, so it came back with nothing, and all the while the data was already sitting right there. I run the NAV database and the open API on top of it, and the agent had no route to either of them.\n\nIt has one now. You give it a fund name, roughly the way you would say it out loud, and it searches the AMFI scheme master and resolves the real scheme code before it fetches anything. If the name could mean more than one scheme it shows you the options and asks you first, since most funds run as Direct and Regular plans with Growth and IDCW options, four near identical names sitting on different series in the data. Several funds in one message is fine.\n\nFor a short window it puts the NAVs in the chat. Ask for longer and it gives you a CSV link rather than printing thousands of rows at you.\n\nIf you would rather do the work yourself, MFPRO (tigzig.com/mfpro) is the interactive tool, go to the \"Data\" tab there, and the MF NAV API (tigzig.com/apis/mf-nav) is open. No signup and no key required for any of them.\n\nLink: quants.tigzig.com\n\nIt does more\nPull prices, company profile and financial statements for anything listed with a Yahoo Finance symbol\nFind a ticker by name, if you know the company but not the symbol. It is now connected to the Yahoo Finance search API\nRun a quant report on one security against a benchmark, covering returns, drawdowns, volatility and the usual risk ratios\nRun a technical analysis report on daily and weekly data, and hand you both the HTML and a PDF\nCompare several securities side by side on return and risk, with the underlying data as CSV\n\nIt can generate all sorts of reports for you, but one thing it is not allowed to do is give investment advice, as that goes into a tricky compliance area.\n\nWhy it asks instead of picking for you\n\nThe same fund exists as Direct and Regular plans with Growth and IDCW options, and they are separate series with different numbers.\n\nPick the wrong one for a return calculation and the answer looks perfectly reasonable and is off by roughly the expense difference every year, compounding.\n\nThis is why the agent shows you the candidates and asks rather than picking for you.\n\nFor example, try this:\n\n``\nshow all schemes of kotak elss\n``\n\n\nRelated\n\nTools: Quants Agent, MFPRO - Mutual Fund Analytics, QRep - Security Reports\n\nExplore: MF NAV API docs, API and MCP catalog, Market and macro tools\n\nMore posts: Three Upgrades on the India Mutual Fund NAV API. Easier to Pull, Whatever You Already Use., Running Hundreds of API Calls to Pull NAV History, One Scheme Per Call? Batch 50 Into One., Pulling Yahoo Finance Data Into Your Own Trackers. Excel, Sheets, a Web Tool, or Straight From an AI Agent., An ISIN Checker for When a Mutual Fund Lookup Fails. And Why One Fund Carries Ten Identifiers., MF NAV API FAQ. Why Am I Getting an Error, Why Is There No SIF NAV, and Why Does the AUM Look Wrong?"
},
{
"slug": "portfolio-agent-symbol-lookup-sep2026",
"title": "The Portfolio Agent Now Looks Up Live Symbols for Stocks and Securities, Instead of Guessing From Memory.",
"tags": [
"portfolio-quants",
"mutual-funds"
],
"content": "The Portfolio Agent Now Looks Up Live Symbols for Stocks and Securities, Instead of Guessing From Memory.\n\nFollowing the mutual fund search I posted earlier, the same agent now does a similar thing for stocks and securities. It looks up the live symbol on Yahoo Finance instead of guessing from memory, and then runs the reports you need: security performance, QRep multi-security, technicals, pulling NAVs and prices.\n\nLog review showed this to be a point of friction, and also that users were asking it to pull NAVs, so I upgraded the agent a bit.\n\nquants.tigzig.com\n\nThe agent runs a bunch of reports for you\nQRep Report. Performance and risk report for one security against one benchmark.\nTechnical Analysis Report. Chart-based technical report for one security.\nSecurity Performance Report. Compare multiple securities over a period, with HTML and CSV output.\nYahoo Finance Data Pull. Prices, profiles, financials, holders, estimates, dividends, splits, actions, and ticker lookup.\nIndian Mutual Fund NAV Data. Scheme lookup and NAV history for Indian mutual funds.\n\nJust share which tickers or mutual funds you want the reports for. The agent will also guide you on choosing the right report.\n\nThe earlier post on the mutual fund side of this: tigzig.com/post/portfolio-agent-mutual-fund-nav-aug2026\n\n\nRelated\n\nTools: Quants Agent, QRep - Security Reports, MFPRO - Mutual Fund Analytics\n\nExplore: Market and macro tools, API and MCP catalog, Claude for Analytics hub\n\nMore posts: My Portfolio Agent Could Not Look Up an Indian Mutual Fund NAV. Now It Can., Pulling Yahoo Finance Data Into Your Own Trackers. Excel, Sheets, a Web Tool, or Straight From an AI Agent., Three Fixes on the Yahoo Finance API This Week, All of Them Found by Reading My Own Error Logs., Three Upgrades on the India Mutual Fund NAV API. Easier to Pull, Whatever You Already Use., Getting Errors on the Tigzig APIs? Is It You, or Is It Me? Now You Can Just Go and Look."
},
{
"slug": "power-up-with-gen-ai-query-analyze-youtube-videos-with-google-notebooklm",
"title": "POWER UP WITH GEN AI: Query & Analyze YouTube Videos with Google NotebookLM.",
"tags": [
"converters-tools"
],
"content": "POWER UP WITH GEN AI: Query & Analyze YouTube Videos with Google NotebookLM.\n\nNotebookLM is proving to be a powerful research assistant, allowing you to query across multiple data sources: PDFs, text files, markdown files, web URLs, copy-pasted text, audio files, and now, YouTube videos.\n\nIn my last post, I showed how you can analyze multiple PDFs of 10Q quarterly filings using Google's NotebookLM.\n\nNow sharing a few use cases for querying YouTube videos-earnings call summaries, detailed transcripts, step-by-step guides, specific questions, and more.\n\n\nRelated\n\nTools: Any File to Text (Markitdown), PDF to Text (Llama Parse), Quants Agent\n\nExplore: MCP & AI Agent Tools, Custom GPTs Hub, All Guides\n\nMore posts: Analyze PDF with NotebookLM. Visualize with Napkin AI., How to use Llama Parse to convert PDF to text and extract complex table data. For Annual Reports, 10Ks, Research Reports, Flowise is my goto platform for GenAI and LLM apps"
},
{
"slug": "powerbots-supercharge-your-business-with-no-code-ai-chatbots-a-practical-guide",
"title": "POWERBOTS : Supercharge Your Business with No-Code AI Chatbots. A Practical Guide",
"tags": [
"custom-gpt"
],
"content": "POWERBOTS : Supercharge Your Business with No-Code AI Chatbots. A Practical Guide\n\nAI Chatbots Guide\n\nAI Chatbots = Power Bots\n\nInsights from hands-on experience\n\nUse Cases: Chat. Process. Automate. Act\n\nAI-enabled chatbots connected to your data sources that can answer questions, process data, automate tasks, and take action. 'Chatbot' is probably a misnomer for today's AI Bots. Chatting is just one function they can do. They are more like business power tools, with potential to supercharge your business. I call them PowerBots\n\nPower Bots: Their integration with Zapier and Make.com allows them to take actions like: update record in leads CRM, qualify incoming leads, set appointments with calendar events, send out email, make and receive phone calls / SMS. And connecting them to an API server, allows them to process and analyze data based on chat instructions and return back results (See my live example of YFIN Bot below)\n\nThere are another 200K+ (estimated) examples in GPT store. Legal Advisors, SEO Analysts, Lead Generation Agents, Web Designers, Adventure Planners, Resume Makers, Job Application Wizards ....covering areas like: health, law, education, productivity, travel, adventure, web design, plant care, books, food, jobs.\n\nAnd yes, their 'regular' chatbots function for customer service, company policy, document Q&A, lead generation, etc., connected to your data sources: PDFs, Text, Tables, URLs.\n\nIn this article, I am looking at no-code custom chatbot builders that allow for rapid build and deployment. Fast-evolving space with options available across a range of requirements. Point to note: some use cases may be beyond the capabilities of no-code builders, but platforms like Flowise offer great options to integrate AI solutions with fully coded platforms.\n\nSummary of Insights:\n\nClassifying into four categories:\nPure Play Chatbot Builders\nChatFlow based\nOpen AI Assistant - Wrappers\nLLM (Large Language Model) Apps Development Platform.\n\nDiscussing 4 chatbot builders in this post covering the four categories: Botsonic / Zapier / Botpress / Flowise. This is a curated and opinionated list, based on my experiences. Many other great platforms out there.\n\nSelecting the right platform would depend on the specific use case. Check out the article to help assess.\nEach of these chatbots can manage simple use cases with equal ease. They are all no-code GUI-based builders.\nFor the simplest and fastest deployment: Botsonic and Zapier.\nIf you need to incorporate multiple forms, interfaces, pages: Zapier.\nIf chat flow-management functionality is required: Botpress.\nFor complex use cases : Flowise.\nTo publish your GPTs on GPT store to web: Flowise / Botpress\nCosting, functionalities, limitations, security considerations, learning curve for tool, ease of use, indicative speed to deploy, and other considerations covered in the article below.\n\nNo-Code AI Chatbot Builders Comparison\n\nNo-Code AI Chatbot builders: Overall Comparison\n\nBasis of Article\n\nThis article is based on my firsthand experience with these platforms, data from official platform resources as well as information from other credible web sources. It includes factual details, insights and opinions based on these experiences and information, as well as my experience with Gen AI technologies. Please note that not every feature mentioned has been individually tested by me, but I have made every effort to ensure accuracy and reliability.\n\nLive App & Prototypes\n\nYFIN bot, developed with Flowise, is live on my website tigzig.com. This live chatbot demonstrates capability of a bot to take instructions, carry out data processing and return back results.\n\nYFIN bot extracts financial data from Yahoo Finance, including P&L, Balance Sheet, Cash Flows, and Quarterlies for multiple periods. It uses Flowise custom functions to make an API call to a custom-built FastAPI server running my Python code.\n\nAs part of live testing, I have also published prototype chatbots on my website for Botsonic, Botpress and Flowise. For this, I took a simple document Q&A use case.....Warren Buffett's 2023 letter shareholders uploaded to knowledge base. Around 21 pages, PDF converted to text. Zapier was only partially tested on their platform itself, as their free tier doesn't allow document upload / website embeds.\n\nDifferent categories for different requirements\n\nPure Play Chatbot Builders\n\nFastest to deploy. Easiest to use. Simple chatbots live in <30 minutes. Fully managed. No Code.\nPlayers: Botsonic, Zapier, Dante-AI, ChatBase, ChatSimple, and others.\nSuitability: Super fast deployment. Very easy to use. For relatively simpler use cases.\nLimitations: Limited customizations.\n\nChatflow-based\n\nTraditional chatbots integrating with LLM Models like ChatGPT.\nPlayers: Botpress, Voiceflow, and others.\nSuitability: Use cases requiring chat flow-management options.\nLimitations_: UI will have a learning curve, though relatively easy to get upto speed. Relatively limited AI specific customizations.\n\nOpenAI Assistant API Wrappers:\n\nWrapper for OpenAI Assistant. Setup a custom GPT/Bot in Open AI Playground and publish on the web via ready to use templates. Deploy in <1 hour. Simple to use. Full power of ChatGPT, so to say, as no RAG/ processing layer in between.\nPlayers: Flowise, Botpress, and others.\nSuitability: To publish your GPT on GPT store to web. Not all GPTs can be 'migrated' easily though. Depends on GPT configuration.\nLimitations: Deploying a GPT outside GPT store increases the cost, sometimes exponentially, depending on the GPT configuration....especially if the GPT uses ChatGPT4. Reason being: for Assistant's Open AI charges based on tokens, and token counts can run into millions very quickly for many uses cases. Pricing (openai.com)\n\nLLM Apps Development Platforms\n\nPlatforms like Flowise offer huge customization, no data limits, custom code support, API calls, Retrieval Augmented Generation (RAG) with Langchain & LlamaIndex, connectivity to vector databases and RDBMS. Deploy in <1 hour. Offer great options to integrate AI solutions with fully coded platforms, in case of highly complex apps.\nPlayers: Flowise, Dify, and others.\nSuitability: From simple to fairly complex LLM applications.\nLimitations: UI will have a learning curve, though relatively easy to get upto speed. No option to create custom frontend / GUI. Flowise needs to be self hosted.\n\nFactors to be considered and comparisons.\n\nFully Managed Vs. Self Hosted\n\nFlowise is Open Source and needs self-hosting, while all the rest are fully managed platforms. See costing details in the section below. Can be hosted on any cloud platform of your choice: Render, Railway, AWS, GCP, Azure. If you are starting up, I would suggest Render for ease of use. Render deployment just needs Flowise GitHub repository to be forked, then connect to Render, choose your plan and disk size, add a few details and that's it. The whole process takes less than an hour.\n\nQuality of Answers & LLM Choice\n\nFor simple use cases, there's unlikely to be much difference in the quality of answers across platforms. However, as the number and complexity of data sources increase, the quality of answers is likely to vary and would need to be tested out.\n\nThe quality of answers depends on the LLM being used, RAG setup (if used), and factors like token size limits if configured.\n\nBotsonic provides LLM choices including GPT-4, Mistral, Google Gemini Pro, and Command R, while Zapier offers ChatGPT-4. Flowise offers whole range of LLMs, while Botpress currently only supports GPT-3.5 and 4 models.\n\nGPT-4 would provide better quality answers, but it is expensive (see LLM cost paragraph for more info). Free tier models like Gemini Pro and Groq (Mistral) are great for many day-to-day use cases.\n\nFlowise provides huge configuration options, including the choice of LLM model and RAG architecture.\n\nCosting\nBotsonic and Zapier have fixed pricing, with add-on options for various functionalities. But pricing is much higher if you use GPT-4 / Mistral large. See LLM cost section below.\nBotpress bots are fixed price, but LLM/ChatGPT costs would be extra.\nFlowise is open-source and therefore free, but LLM and server costs would be separate.\nServer costs for Flowise on Render: The Starter pack costs $7 per month for 512MB RAM and 0.5 CPU, with support for up to 512GB RAM and 64 CPUs. Additional disk space at $0.25 per GB per month. The Starter pack and 1 GB are enough to start off with for testing. Render has a great free tier if you just want to check out Flowise, but otherwise the free tier is not feasible for Flowise. With the free tier, the server goes to sleep after half an hour of inactivity and takes around a minute to 'wake up' and also you lose your saved chatbots, though you can export them to save locally and reimport later.\nLLM cost, for both Flowise and Render, can vary depending on the LLM you choose. ChatGPT-4, though very powerful, is also very expensive at around $30 per million tokens (mtk) for output and $10/mtk for input. In comparison, ChatGPT-3.5 is $1.5 mtk for output and $0.5 mtk for input. Google Gemini Pro has a great free tier with a rate limit of 60 rpm (requests per minute). Groq is currently free, with rate limits of 40 rpm, 40K tokens per minute, and 14.4K requests per day, and I believe it uses one of the Mistral models in the integration with Flowise. Other models' pricing and free tiers would vary. Flowise also provides access to open-source models.\n\nPricing Comparison\n\nNo Code AI Chatbot Builders: Pricing Comparison\n\nSecurity and Risk of Proprietary Data Leakage\n\nThis is a known issue with all LLM Apps, including chatbots. For unsecured chatbots, with simple prompts, the bots will share all custom instructions as well as file content from uploaded files.\n\nBotsonic offers protection against basic hacks. I didn't check for more advanced hacks. Zapier didn't seem to have basic protection built-in and would need to be incorporated. Botpress and Flowise are more customized solutions and would need security measures to be incorporated separately.\n\nSecurity Features Comparison\n\nNo-Code AI Chatbot builders: Comparision of Security Features\n\nAutomation and Actions with Zapier & Make\n\nConnecting a chatbot to Zapier/Make allows a chatbot to 'take action'. For instance based on information collected during a chat, the chatbot can update a leads database / google sheets / CRM. Or a chatbot can potentially be automated to trigger an inbound or an oubound call (see link at end).\n\nZapier and make connect to thousands of platforms allowing huge potential for such actions and automations. YouTube has some amazing examples. Sharing a short list in resources section at the end.\n\nCustomization Options\n\nAdditional Functionalities\n\nNo-Code AI Chatbot builders: Additional Functionalities\n\nBotsonic and Zapier are pre-configured chatbots with limited customization options, but they do offer a fair degree of functionality. One great feature with Zapier is its interfaces, forms, and pages. Botpress provides a good range of customization options, particularly for configuring chatflows. Flowise offers the highest level of customization as it is a full-fledged LLM Apps development platform. Complex customization is relatively straightforward in Flowise.\n\nAdvanced Features\n\nNo-Code AI Chatbot builders: Comparision of Advanced Features\n\nLarge data corpus / RAG\n\nIf you have a large corpus of documents, then it might need an RAG setup. This video from IBM provides a nice and simple explanation of RAG: What is Retrieval-Augmented Generation (RAG)?\n\nRAG architecture and optimization are still evolving areas. Whether to use RAG or not, and if so, what kind of architecture to use and how best to optimize is a whole different area in itself.\n\nBut if you do need to use RAG, Flowise is probably the best choice given that it's based on the Langchain framework and also supports Llama Index, both with great RAG solutions.\n\nAt the same time, do keep in mind that RAG may not always be required or necessary. With increasing context size windows and decreasing LLM costs, many use cases can do without RAG. Here's a nice video from Lyzer AI. Argues for RAG, but explains concept very nicely: Is RAG Dead?\n\nKnowledge Base Functionalities\n\nNo-Code AI Chatbot builders: Knowledge Base\n\nRapidly Evolving Field\n\nGen AI space as well as chatbot builders is a rapidly evolving field. Traditional chatbot builders like Botpress and Voiceflow are rapidly integrating with Gen AI technologies. In fact, Botpress is moving towards what it calls a 'GPT Native' platform. Platforms like Flowise are adding features by the day. New players like Zapier are coming in . On the LLM side, LLM costs are dropping, context windows are increasing, free tiers are getting larger and more and more 'high quality' open source models are coming into play. And deployments as well as integrations are getting easier and cheaper.\n\nAnd this is just the beginning.\n\nAs the whole Gen AI space evolves, very likely the landscape would look very different a year from now.\n\nCommunity Support, References and Links\n\nThere are lots of excellent video tutorials on YouTube on AI Chatbots, covering functionalities, usage and applications. Very grateful to community members for sharing such valuable information. Below is a small list.\nFlowise Tutorial from Leon van Zyl. Excellent tutorial. Makes getting upto speed on Flowise fast and easy Flowise AI (2024) Tutorial - YouTube\nFlowise Tutorials: Thomas Ingram has some excellent videos on Flowise apps: Thomas Ingram - YouTube\nBotpress Channel on YouTube has great videos to help get upto speed: Botpress YouTube Channel\nBotpress GPT-native platform: blog releasing its new GPT-native platform, along with a Video guide to new platform. Unveiling the new Botpress: a GPT-native bot-building platform | Botpress Blog\nLink to Botpress template for rapid deployment of Open AI Assistant: How To Deploy a Custom OpenAI Assistant to Your Website or Messaging Channel | Botpress Blog\nHow to build a classic lead capture chatbot in Botpress by connecting to Zapier and Make. Can be used for any chatbots that can connect to Zapier / Make / Webhook. Video from Alex Make | Automation Software | Connect Apps & Design Workflows\nEnable chatbot to make or receive call: Amazing video and receiving and making calls. This is slightly different use case, but I believe the framework can be utilized in any chatbot that can link up with Zapier/ Make. Sending and Receiving a call to chatGPT using Whisper, Twilio & Zapier - YouTube\nEnable chatbot to send personalized SMS: Excellent video from Corbin on automation with Zapier and ChatGPT. Here the trigger is a Mailchimp link click which ultimately sends out a personalized SMS. This can easily be adapted to use with chatbot by just changing the trigger to an action taken by user in your chatbot. Zapier and ChatGPT For Twilio: OpenAI For Making SMS Text | Tutorial - YouTube\n\n\nRelated\n\nTools: YFIN Bot, ChatGPT via n8n, Quants Agent\n\nExplore: Custom GPTs hub, MCP Hub, Browse all apps\n\nMore posts: Flowise is my goto platform for GenAI and LLM apps, LLM App | FastAPI Server | Web, Quick Deploy Advanced Analysis Multi-Agent with Flowise, Code Red: Unprotected GPTs & AI Apps exposed by simple hacks"
},
{
"slug": "private-credit-bdc-stress-regulators-jun2026",
"title": "Private Credit and BDC Stress. Discounts at 13-Year Lows. Withdrawals Past the Gate. Regulators Now Arriving.",
"tags": [
"portfolio-quants"
],
"content": "Private Credit and BDC Stress. Discounts at 13-Year Lows. Withdrawals Past the Gate. Regulators Now Arriving.\n\nPublished: 2026-06-06\n\nThe median listed BDC trades at 25% discount to NAV, near its deepest in 13 years excluding Covid - 0.75x price to book today. Pimco's CIO: 'the first sustained default or loss cycle in many, many years.' Q1 losses: FS KKR $441M, Blue Owl's BDC $184M. Withdrawal requests blew past the 5% redemption gate at Blackstone's BCRED (~10%) and Cliffwater (17%) - a first-ever net outflow. Private-credit rating downgrades up 50% in a year. UBS sees defaults rising from ~4.4% to 9-10% by end-2026. US issuance fell ~40% vs Q1 to $44.8B in the 3 months to May. Partners Group warns the redemption pressure that began in private credit is now spreading into private equity. The regulators have started arriving: federal prosecutors in NY questioning BlackRock executives over how their credit fund valued loans; FSB flags private credit's leverage and interconnections as a source of broader risks to financial stability; ECB puts US private credit on its watchlist as a spillover risk into Europe; Sarah Breeden at the Bank of England calls it a potential 'market for lemons' (Akerlof). Companion update to the prior 'Is Private Credit Turning Into a Lemon' analysis.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/private-credit-bdc-stress-regulators-jun2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/private-credit-bdc-stress-regulators-jun2026 with header Accept: text/markdown\n\nTags\nportfolio-quants\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QRep - Security Reports, Quants Agent\n\nExplore: Private Credit & NDFI Tracker, Quant Apps hub, Analysis hub\n\nMore posts: Private Credit Default Rate Hits a Record 6.0% (2026): A Market for Lemons? Banks and Insurers Deep In It., The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top., Private Credit - The $2.7 Trillion Shadow Lending Market Is Showing Cracks, FSB Just Published 50 Pages on Private Credit. The Title: 'Vulnerabilities in Private Credit.' Worth Reading., Banks Holding the Line.. But the 'Lemons' Sit Off the Chart. In Private Credit, Credit Unions, and Non-Bank Consumer Lenders."
},
{
"slug": "private-credit-market-for-lemons-may2026",
"title": "Private Credit Default Rate Hits a Record 6.0% (2026): A Market for Lemons? Banks and Insurers Deep In It.",
"tags": [
"portfolio-quants"
],
"content": "Private Credit Default Rate Hits a Record 6.0% (2026): A Market for Lemons? Banks and Insurers Deep In It.\n\nPublished: 2026-05-28\n\nPrivate credit's default rate hit a record 6.0% in April 2026 (Fitch, trailing 12 months). Bank of England's Sarah Breeden invokes Akerlof's 'market for lemons'; a government report calls it a layered cake like 2008 subprime. Banks and insurers are deep in - JPMorgan $22B, Citi $22B, Deutsche $30B - with a 19% off-cycle NAV writedown, a $648M bank credit-line cut, KKR rescuing its own FSK fund, and a DOJ probe into BlackRock TCPC valuations. Numbers with sources.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/private-credit-market-for-lemons-may2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/private-credit-market-for-lemons-may2026 with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/PRIVATECREDITMAY2026.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nTIGZIG · Private Credit · The May 2026 Update\n\nBanks and insurers are deep in it.\n\nSarah Breeden, Deputy Governor for Financial Stability at the Bank of England, warned this April that private credit could tip into a \"market for lemons\" - George Akerlof's Nobel idea, where information asymmetry and adverse selection drive out the good ones, leave the bad, and the market can collapse.\n\nSame pattern as 2008 - subprime mortgages turned into lemons.\n$1.5-2T - market size (FSB).\n6.0% - record default rate, April (Fitch).\n\nThe Story So Far\n\nThis is the fourth dispatch in the series. If you are new, here is the gist of the prior three.\nDispatch 1 · 24 Mar - The $2.7T Shadow Lending Market Is Showing Cracks. Defaults forecast near the Covid peak. Listed managers down 30-50%. Dimon: \"when you see one cockroach, there are probably more.\"\nDispatch 2 · 26 Mar - US Banks & Non-Bank Lending. How Deep is the Exposure? $1.57T of US bank loans sit behind the non-bank lenders - the banks fund the people making the risky loans.\nDispatch 3 · 8 May - Red Flag in US Life Insurance. Up to $2.4T exposure. PE often owns both sides - the insurer and the credit fund - and decides what the loans are worth.\n\nAcross banks, non-banks and insurers - nobody has the exact number, because so much of it is undisclosed and layered. The stress is real and rising fast - and it will be felt across banking and insurance.\n\nWhat Changed in Eight Weeks\n\nFor two years this was a slow drift - in eight weeks it became a cluster.\nApr 2026 - IMF. Global Financial Stability Report flags private credit - selective defaults could run 2-3x higher under stress.\nApr 7 - Moody's. Cuts the US BDC sector outlook to negative from stable.\nApr 2026 - Bank of England. Running industry stress tests. Breeden's \"this time is different?\" speech.\nMay 6 - Financial Stability Board. Publishes its Vulnerabilities in Private Credit report - data gaps, valuation, opacity.\nMay 7 - Office of the Comptroller of the Currency. Warns PIK and restructurings \"may mask underlying credit deterioration.\"\nMay 11 - FS KKR (FSK). Breaks in public - banks cut the credit line, KKR steps in days later.\nMay 17 - Department of Justice. Probing valuation marks at a BlackRock private credit fund.\nMay 18 - Fitch. Private credit default rate hits a record 6.0% (April TTM) - the highest since its index began.\nMay 27 - European Central Bank. Financial Stability Review names US private credit as a spillover risk to the euro area.\n\nThe Data Turned\n\nThe default rate just hit a record - and it has been climbing every month.\nDec 5.6% → Apr 6.0% - Fitch private credit default rate (trailing 12 months). A record high, climbing every month into April.\n4.8% (MCO) - the CLO book (~1,000 issuers).\n9.7% (PMR) - the insurer book (~300 issuers).\n\nThe two books split sharply: the insurer-facing PMR book runs at nearly double the MCO CLO book - and PMR actually eased a touch, from 10.0%.\n\nA note on the method: all rates are trailing 12 months, and Fitch counts by number of defaulters, not dollars. That cuts both ways - a few large names can't inflate the rate, but it also tells you nothing about the dollar scale. Nine small defaults and nine large ones read the same here.\n\nSources: Fitch, US Private Credit Default Rate, May 18; Fitch PCDR Methodology.\n\nThe Measurement Fight\n\nThe default number you believe depends on what you choose to count.\n\nFitch is clear about what counts as a default. Of the events in April, here is what they actually were:\n55% - PIK / interest deferral. Borrower stops paying cash interest and rolls it into the loan.\n35% - Maturity extension under stress. The due date gets pushed out to avoid a miss.\n6% - Bankruptcy or liquidation. The company genuinely goes under.\n\nSo roughly 9 in 10 \"defaults\" are stress restructurings, not companies dying. Fitch labels them defaults anyway - and it is right to.\n\nMoody's reads it the same way: about 65% of 2025 private credit defaults were distressed restructurings, swaps and extensions agreed under duress.\n\nThat is the whole fight. Count the restructurings and the rate looks alarming. Strip them out and it looks calm. Same loans - two very different stories.\n\nSources: Fitch PCDR April 2026; Moody's, US corporate default risk 2026.\n\nA KKR Fund Breaks in Public\n\nWhen even one of KKR's own funds needs rescuing - you read the order of events.\nMay 8 (Fri): JPMorgan-led banks cut the fund's credit line by $648M, raised rates, and some lenders walked away entirely.\nMay 11 (Mon): KKR steps in with a $300M rescue of its own fund - fresh equity, plus cash to buy out investors heading for the door.\n8.1% of the fund's own loans have stopped paying, up from 5.5% at year-end. A Q1 loss of roughly $560M already.\n\nThe tell. FSK is a $12.3B fund co-managed by KKR, one of the biggest names in private credit. Its biggest exposure is software - the sector everyone is worried about. Shares down nearly half this year. Moody's cut it to junk in March.\n\nIf a fund run by KKR needs a rescue - you have to wonder what the smaller, less-backed funds look like. And the banks have only pulled their line so far - no writedown disclosed yet.\n\nSource: CNBC, FS KKR - JPMorgan-led group reins in credit, May 11.\n\nBlackRock TCPC and the DOJ\n\nWhen there's no market price, the mark is an opinion - and opinions are now under federal scrutiny.\n\nTCPC filed a rare off-cycle disclosure in January - NAV cut 19% in one go ($8.71 to $7.07). Stock fell 13% that day, the most since March 2020 - and is down 24% on the year.\n\nThe big one: class-action suits followed, claiming the fund made \"materially false\" statements and didn't properly value its loans. The Manhattan US Attorney (SDNY) is now seeking information and has questioned executives over valuation practices.\n\nJay Clayton, SDNY: \"if people are mismarking in order to generate fees, that's always been a no-no.\"\n\nIn a market with no daily price, the NAV is the manager's own number - that is the whole model, and it is now the thing under investigation.\n\nSource: Bloomberg, BlackRock private credit fund's valuations are probed by DOJ, May 17.\n\nBanks - The Back Door\n\nBanks left the lending - they never left the financing.\n\nAfter 2008, regulators pushed banks out of risky middle-market lending. They came back as the financiers behind it - credit lines, warehouse facilities, leverage. By late 2025 US banks had extended nearly $300B of credit to private credit funds, BDCs and CLOs (Moody's).\n$22B - JPMorgan direct exposure - the same bank now leading the pullback at FSK, and whose CEO is warning loudest.\n17% - Wells Fargo - share of its $36B corporate book tied to software, the sector under the most pressure.\n$22B - Citigroup - no losses reported yet, but flagging \"active monitoring.\"\n\nJPMorgan's Jamie Dimon warned in his 2026 letter that losses will be \"higher than expected\", hitting the industry's lack of \"rigorous valuation marks.\" The Fed has since formally queried major banks on their exposure.\n\nAnd it is not just US banks - European disclosures this season ran $20-30B each (Deutsche $30B, BNP $25B, Barclays $20B). Deutsche's disclosure alone dented its share price.\n\nSources: Forbes, May 24; CNBC, May 6.\n\nInsurers - The Quiet Channel\n\nThe loudest stress is in the funds - the quietest is on insurance balance sheets.\n\nPrivate credit held by US life insurers grew 20%+ in 2025 - now ~10% of total assets, and over 15% for PE-affiliated insurers like Athene and Global Atlantic.\n\nThe FSB warns these PE-insurer structures may \"exploit flexibility and opacity in valuations - regulatory arbitrage - and offshore reinsurance jurisdictions.\" A regulator's words, not mine.\n\nThe US Treasury has assembled a team to assess insurer exposure and will meet state regulators on the emerging risk.\n\nWhy it's quiet. Insurers mark to model and hold to maturity - losses surface slowly, if at all, in public filings. The IMF warns that insurers holding complex, leveraged private credit, often rated investment grade with thin capital buffers, could face larger-than-expected losses in a stress.\n\nThe fund stress makes headlines. The insurer exposure won't - it sits in slow-moving annuity books that only reveal stress years later, by which point retirees are already holding it.\n\nSources: Forbes, May 24; FSB, Vulnerabilities in Private Credit; IMF GFSR, Apr 2026.\n\nThe Layer Cake - What Makes This Hard to Size\n\nThe risk isn't one bad loan - it's leverage stacked on leverage, the way 2008 was.\n\n\"Leverage is a layer cake - at the borrower, fund and sponsor level - making it hard to measure. Loans are sometimes sliced and diced into CLOs - losses that are already hard to size become even harder to trace.\"\nSarah Breeden, Deputy Governor, Bank of England · \"This time is different?\"\n\nThe FSB says the same: borrower-level debt is driven by PE sponsors buying companies with \"a high proportion of debt\" - then the credit funds lend on top, the funds borrow on top of that, and reporting captures only \"the first layer.\"\n\nLeverage piled three deep and sliced into CLOs is exactly the shape 2008 took.\n\nSources: Bank of England, Breeden speech; FSB, Vulnerabilities in Private Credit.\n\nThe Layer Cake (Diagram)\n\nLeverage stacks up - funding feeds in - the risk gets repackaged out. All of it balances on one borrower.\n\nAbove the line - Funding (who puts money in): Banks & NBFIs, Insurance, Retail (via BDCs), Pensions & family offices. Sponsor (PE firm, borrows at the top). Fund (borrows to lend more).\n\nRepackaging - sliced and sold on: CLOs - loans pooled into rated tranches - then bought by banks, insurers and other funds. The same names on the left.\n\nBelow the line - the borrower: the company that actually owes. When the borrower defaults, the whole deck of cards above it collapses.\n\nA simplified picture - the real market is far more fragmented, with many more players and sub-segments, and linkages that run diagonally and loop back on each other in ways a clean diagram can't show.\n\nRegulators (1 of 2) - International\n\nThe data isn't good enough to see the risk. Four bodies, four jurisdictions, one message.\nFSB · May 6. Wants national regulators to step up supervision - flags patchy loan-level data, valuation subjectivity, reliance on small private-rating shops, and \"circles of risk\" where banks fund the funds that buy banks' own risk transfers.\nIMF · April. Liquidity mismatch looks contained to semi-liquid structures for now. But under a stressed-rates or stressed-earnings scenario, selective default rates could rise two to three times.\nBank of England · Breeden, April. Running stress tests alongside the industry. \"This time is different?\" - risk migrates to the opaque, least-regulated corners, and crises hit when several vulnerabilities crystallise at once.\nECB · Financial Stability Review, May 27. Names US private credit as a spillover risk to euro area stability. Flags complex leverage structures, opaque valuations, PIK concentration and AI-disruption risk at US BDCs.\n\nSources: FSB Report; IMF GFSR Apr 2026; BoE Breeden speech; ECB FSR May 2026.\n\nRegulators (2 of 2) - US Supervisors\n\nFrom watching, to a federal probe. In months, the US regulators went from monitoring to questioning to a prosecutor's probe.\nOCC · Semiannual Risk Perspective, May 7. PIK and debt restructurings \"may mask underlying credit deterioration\" in private credit portfolios. Flags rising bank concentration to these funds as a monitoring priority.\nFed. Has formally queried major banks on their private credit exposure - a clear posture change from passive monitoring to active questioning.\nDOJ · Manhattan US Attorney (SDNY). Federal prosecutors are scrutinising the valuation marks at a BlackRock BDC, with executives questioned. And Clayton has said publicly that \"mismarking to generate fees\" has \"always been a no-no\" - the first time prosecutors have leaned on how private credit sets its prices.\n\nSources: OCC Semiannual Risk Perspective; Forbes, May 24; Bloomberg, DOJ probe.\n\nThe Counterview - The Bulls Have a Case\n\nThey're not ignoring the stress - they're reading a different slice of the same data.\n\nS&P Global · March. \"Fundamentals remain resilient.\" Median interest coverage up to 1.84x from 1.74x. Upgrades roughly equal downgrades (65 vs 63). Default rate including selective defaults actually fell to 4.37%.\n\nGoldman panel · Apr-May (Lynam · Arougheti, Ares). Redemption gates are a feature, not a bug - designed to prevent fire sales. The scary retail BDC slice is only ~15% of private credit - the rest is locked-up institutional money. Realized losses below historical average. \"No run on the bank.\"\n\nThe industry's line, via the American Investment Council: \"The private credit system is working as designed.\"\n\nSources: S&P Global; Goldman Sachs Exchanges, \"Cracks in Private Credit\"; CNBC.\n\nThe Bear View - Gundlach: \"A Little Bit Like 2006\"\n\nThe DoubleLine founder called the subprime collapse early - now he's pointing the same finger at private credit.\n\n\"A little bit like 2006, where everything is overvalued, cracks are starting to form. But everyone's like, it's all contained - it's just software. But it's not just software.\"\n\nHe says private credit has \"the same trappings as subprime mortgage repackaging had back in 2006\" - and calls it the next big crisis in the markets.\n\nOn the valuation problem: private assets really have two prices - 100 or zero. Marked at par one week, near-worthless the next when the borrower files (he points to Renovo).\n\nA warning on timing: being right is not the same as being early. Being negative on mortgages in 2004 took three years to pay off.\n\nSources: CNBC, Mar 2026; Fortune / Bloomberg Odd Lots, Nov 2025.\n\nMy Read - It Always Looks Contained, Until It Doesn't\n\nIn 2008 the cracks were already showing - losses creeping up, indicators rolling over. But the line everybody parroted was the same one, over and over: no systemic risk, the system is resilient, these are minor stresses, cycles come and cycles go.\n\nThat line held right up until Lehman. Then the tone flipped overnight, and every skeleton fell out of the cupboard at once.\n\nWhat is different this time is not the rhetoric - that is identical. It is that I can see many indicators cracking at once - labor, consumer delinquencies at or past 2007 levels, the layered credit in this deck, an S&P running on its own exuberance.\n\nUsually one thing cracks at a time. Right now it is several, together, while one corner of the world dances as if it is all fine.\n\nAnd there is one thing in here that is new - AI displacing labour. Computerisation, the internet, mobile - each played out over a period of time. This one is playing out instantly. So the pattern is similar to 2008 - but it is carrying a force none of the old cycles had.\n\nI am not calling a crash - nobody can time these things.\n\nBut this is the part of the movie where everyone says it is contained.\n\nUntil...\n\nMore: my deeper work on fundamentals, macros and markets is at tigzig.com/analysis.\n\nSources Index\n\nPrior dispatches (Tigzig):\nPrivate Credit · The $2.7 Trillion Shadow Lending Market Is Showing Cracks (Mar 24)\nUS Banks & Non-Bank Lending · How Deep is the Exposure? Is There a Systemic Risk? (Mar 26)\nRed Flag in US Life Insurance · $2.4T (7.5% of GDP) in Private Credit and Offshore Reinsurance (May 8)\n\nExternal sources:\n[01] Forbes, Rising Private Credit Defaults Are Testing Banks And Insurers (May 24)\n[01a] Fitch, U.S. Private Credit Default Rate Hits a High of 6.0% in April 2026 (May 18)\n[01c] Fitch, U.S. Private Credit Default Rate Methodology (Jan 30)\n[02] CNBC, FS KKR fund: JPMorgan Chase-led group reins in credit (May 11)\n[04] FSB, Report on Vulnerabilities in Private Credit (May 6)\n[05] CNBC, Financial Stability Board sounds alarm on private credit stress (May 6)\n[07] Bloomberg, BlackRock private credit fund's valuations are probed by DOJ (May 17)\n[10] S&P Global, Private Credit Fundamentals Remain Resilient (Mar)\n[12] Goldman Sachs Exchanges, \"Cracks in Private Credit\"\n[13] IMF, Global Financial Stability Report, Chapter 1 (April)\n[15] Bank of England, Sarah Breeden \"This time is different?\" Harvard speech (April)\n[16] ECB, Financial Stability Review, May 2026 (US private credit as spillover risk)\n[17] Moody's, US corporate default risk in 2026\n[18] OCC, Semiannual Risk Perspective, Spring 2026 edition (May 7)\n[19] CNBC, Jeffrey Gundlach warns of private credit strains in a \"going nowhere\" market (Mar 23)\n[20] Fortune, Gundlach on Bloomberg Odd Lots, private credit as the next crisis (Nov 18)\n\nAmar Harolikar - Decision Sciences & Applied AI - tigzig.com - Tools for Analytics, Quants & Macro Signals.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: Private Credit & NDFI Tracker, Analysis hub, Quant apps hub\n\nMore posts: Banks Holding the Line.. But the 'Lemons' Sit Off the Chart. In Private Credit, Credit Unions, and Non-Bank Consumer Lenders., Private Credit - The $2.7 Trillion Shadow Lending Market Is Showing Cracks, Private Credit and BDC Stress. Discounts at 13-Year Lows. Withdrawals Past the Gate. Regulators Now Arriving., Dimon Was Right. More Cockroaches Arrive. UK Lender MFS Collapses - $2.5B Across 12+ Global Banks., FSB Just Published 50 Pages on Private Credit. The Title: 'Vulnerabilities in Private Credit.' Worth Reading."
},
{
"slug": "private-credit-shadow-lending-cracks-2026",
"title": "Private Credit - The $2.7 Trillion Shadow Lending Market Is Showing Cracks",
"tags": [
"portfolio-quants"
],
"content": "Private Credit - The $2.7 Trillion Shadow Lending Market Is Showing Cracks\n\nPublished: 2026-03-24\n\nPrivate credit showing cracks that remind Wall Street of 2008. Jamie Dimon and Lloyd Blankfein flagging risks. 40% of borrowers have negative cash flow. Blackstone scrambled to meet $3.8B in redemptions. Management companies down 30-50% in 6 months.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/private-credit-shadow-lending-cracks-2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/private-credit-shadow-lending-cracks-2026 with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/TREMORMAR2026PRIVATECREDIT.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nPrivate Credit\n\nStraw that breaks the camel’s back?\n\nGhosts of 2008“When you see one cockroach, there are probably more…” Jamie Dimon\n\nWhat is Private Credit & Direct Lending\n\n2025E Private Lending ~2.7T & direct lending ~1.4T\n\nDirect Lending > 50% of all private credit\n\nPrivate credit is lending outside the traditional banking system - private lenders originate loans directly with borrowers\n\nPost-2008, banks retreated from lending to smaller firms. Direct lending filled that gap for middle market companies\n\nBanks are still exposed - they fund private lenders (BDCs)\n\nRetail/HNW investors exposed via BDCs, new private credit ETFs & Pension fund investments\n\nSourcesThe Evolution of Direct Lending - Morgan StanleyPrivate credit investments: 'Some caution is reasonable,' advisor says\n\n“A significant private credit shakeout on par with Covid losses is coming” – Morgan Stanley\n\n“In our view, AI disruption will be a meaningful catalyst to drive defaults higher in direct lending… Overall, we expect the direct lending default rates to reach 8%, approaching Covid peak levels”\nstrategist Morgan Stanley Joyce Jiang (as reported by CNBC)\n\nPremise: New technology & AI ➜ erode demand for software services ➜ hurt lenders. Morgan Stanley estimates that ~26% of direct lender portfolios (BDCs) and ~19% of private credit CLOs are exposed to software - a sector highly vulnerable to AI disruption\n\nSource: Private credit shakeout matching Covid losses coming, Morgan Stanley says\n\nInvestors in private credit demanding their money back\n\nBlackstone, the private equity giant: $3.8 billion in redemption requests. CNBC reports that at least 25 senior leaders from across the firm pitched in some $150 million from their own wallets .\n\nBCRED, Blackstones flagship private credit fund posted its first monthly loss of 0.4% in more than three years in February, it’s first since September 2022, when it posted a loss of 1.3%.\n\nBlue Owl - hit with withdrawal requests. Halts redemptions and liquidates assets to repay.\n\nSourcesBlackstone's flagship private credit fund posts first monthly loss in over three years | ReutersWhy Wall Street is calling out 'echoes' of the 2008 financial crisis | CNN BusinessAfter pouring billions into private credit, many investors want out\n\n“We're due for kind of a reckoning\" Lloyd Blankfein, ex-CEO Goldman Sachs\n\nLloyd Blankfein on Bloombergs ‘The Big Take’ podcast (as reported by Yahoo Finance) while speaking on the troubles in the private credit market.\n\n“The market could be the flashpoint for a bigger crisis in markets, pointing to the growing risks of another financial crisis given the decade of relative calm since the 2008 downturn.”\n\n\"It sort of smells like that kind of a moment again,\" Blankfein said, referring to the 2008 financial crisis. \"I don't feel the storm, but the horses are starting to whinny in the corral.\"\n\nSource: Ex-Goldman CEO Sounds Alarm on Private Credit: Could the Next Crisis Be Around the Corner?\n\n“When you see one cockroach, there are probably more… Everyone should be forewarned on this” Jamie Dimon\n\nTricolor Holdings (subprime auto lender) went bust Sep 2025. Even JPMorgan took a $170M hit from this single bankruptcy\nFirst Brands (auto-parts) filed Chapter 11 with ~$2.3B in hidden loans. DOJ opened a criminal probe - echoes of Lehman\n\nDimon’s point: these may be isolated, but in a downturn more will surface - “when you see one cockroach, there are probably more”\n\nBeyond software, are car loans the next ‘toxic mortgages’ ?\n\nSource: Why Jamie Dimon is warning of 'cockroaches' in the US economy | CNN Business\n\nThe regulators saw it coming\n\nIMF April 2024: Private credit could become systemic risk. Stale valuations, fragile borrowers, hidden leverage, opaque interconnections.\n\nIMF 2025: Half of DL borrowers had negative free operating cash. This deterioration of the borrowers’ credit quality has not been reflected in the accounting valuation of DL loans.\n\nBoston Fed May 2025: Banks fund private lenders through credit lines and could pose systemic liquidity risk to banking sector under certain circumstances, including borrowers defaulting together. Boston Fed’s own words \"tail risk may be underappreciated”\n\nSourcesIMF Global Financial Stability Report, April 2024, Chapter 2IMF Global Financial Stability Report, October 2025Federal Reserve Bank of Boston, Current Policy Perspectives 25-8, May 2025\n\nWhat are the markets saying?\n\nBlackstone down ~40% from its 6 month highs\n\nPast 10 Years\n\nPast 6 Months\n\nApollo Global down ~30% from 6m highs\n\nPast 6 Months\n\nPast 10 Years\n\nBlue Owl Capital down ~50% from 6m highs\n\nPast 10 Years\n\nPast 6m\n\nSince IPO – Past 5yrs\n\nAres Management down 40% from 6m highs\n\nPast 1 Year\n\nPast 10 Years\n\n“The US private credit market is heading in the wrong direction” - The Kobeissi Letter\n\nSource: The Kobeissi Letter on X: \"The US private credit market is heading in the wrong direction\n\nBDCs are publicly traded firms that lend to small and mid-sized US businesses.\n\nAt 0.73x NAV, the market is saying 27% of their loan book is overvalued.\n\nThis ratio has been declining for 18 months.\n\nIn 2008 it hit 0.35x.\n\n~20% of BDC portfolios are exposed to software, a sector under AI disruption pressure.\n\nThe macro backdrop makes it worse\n\nStructural weaknesses intensifying\n\n↓\n\nNon Farm\nPayroll\n\n2-yr downtrend\n\n↑\n\nJobless\nRate\n\nCreeping up\n\n↑\n\nBrent\nCrude\n\nNearly 2x\n\n↑\n\nPPI\nCore\n\nRising fast\n\n↑\n\nPPI\nOverall\n\nOil shock pending\n\n↓\n\nRetail\nSales\n\nDeclining\n\n↑\n\n30+ DPD\nDelinquency\n\nPast COVID highs\n\n↓\n\nYield\nCurve\n\nFalling\n\n↓\n\nEquity\nMarkets\n\n5-10% off highs\n\nUnemployment creeping up steadily the past 2 years\n\nPast 20 Years\n\nUS 30+ DPD has already crossed the COVID highs\n\nPast 20 Years\n\nPrevious Analysis going into structural weaknesses, oil prices and market movements:\n\nStructural Weakness Intensifying March 20, 2026\n\nAre We Staring into Face of Another Full BearMarch 14, 2026\n\nYou don’t know who’s been swimming naked until the tide goes out. - Buffett\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QRep - Security Reports, Quants Agent\n\nExplore: Private Credit & NDFI Tracker, Quant Apps hub, Analysis library\n\nMore posts: Private Credit and BDC Stress. Discounts at 13-Year Lows. Withdrawals Past the Gate. Regulators Now Arriving., Private Credit Default Rate Hits a Record 6.0% (2026): A Market for Lemons? Banks and Insurers Deep In It., Dimon Was Right. More Cockroaches Arrive. UK Lender MFS Collapses - $2.5B Across 12+ Global Banks., The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top., Bank Loss Rates Look Calm. The Stuff Underneath - Consumer, Private Credit, Non-Banks - Says Otherwise."
},
{
"slug": "pushover-server-down-alerts-jul2026",
"title": "The boring $5 tool that tells me when a server goes down",
"tags": [
"infrastructure"
],
"content": "The boring $5 tool that tells me when a server goes down\n\nYou run a couple of backends, a database, a frontend, maybe a small server on Hetzner or some VPS. One of them falls over on a Sunday afternoon. How do you find out? Is it the client calling you? Or you happening to open the site? Or catching it in the server dashboard you keep? That is not monitoring. That's luck.\n\nThe piece that is usually missing is a channel that reaches your phone the second something is wrong, through Do Not Disturb if it is serious, without you building an app or paying for a heavy enterprise pager. I did not know a clean answer to this existed until a few weeks ago. It is called Pushover, it has been around since 2012, it costs a one-time $5, and it has quietly become the thing that tells me when a backend dies, a TLS cert is about to expire, or someone logs in from an IP that is not me.\n\nThe one-line version\n\nPushover is a \"send a notification to my phone\" button that any script or server can press with a single web request. Your code makes the request, a notification shows up on your phone a second later. That is the whole product.\n\nHow it actually works\n\nThe flow goes one way, your side makes one outbound call to Pushover, you are never waiting for them to reach into your server. Three hops:\nSomething on your side notices a problem. That something is just code you already control, a cron job that pings your backend every couple of minutes, your app's error handler, a small monitoring script.\nThat code makes one outbound HTTPS POST to Pushover's API, carrying a title, a message, and a priority.\nPushover's servers get it and push it to the Pushover app on your phone. They own all the hard delivery machinery, the device registry, the Apple and Google push plumbing, the retry logic, the sounds.\n\nSo when people say it lives on their server, that is true in the useful sense, all the delivery machinery is theirs and hosted. You never run a notification server, never touch push certificates, never keep a device list. Your only job is the one outbound POST. Picture it as your code, one POST, Pushover's cloud, your phone.\n\nThe one-time setup\n\nThe setup is about ten minutes, done once.\nInstall the app and get your two keys. Install Pushover on your phone and create an account, which registers the phone as a device. Then in the web dashboard create an application. You now have two strings, a user key that identifies you and an app token for the script, and your code sends messages using those two, no OAuth and no SDK, just the values and an HTTPS POST. It is free for a 30 day trial, after which it is not free, a one-time $4.99 per platform covers it (iOS, Android and desktop are billed separately, per Pushover's pricing page).\nFlip one phone-side toggle. To let emergency alerts ring through Do Not Disturb, grant the app notification permission and turn on its critical-alert override, once. That is the permission step people vaguely remember.\nKeep the two keys out of your source code. Put them in an environment variable or a secrets file like any other credential. Mine sit in an env file the cron jobs read, never in git.\n\nA representative snippet\n\nHere is the entire integration, in shell:\n\n``bash\ncurl -s -X POST https://api.pushover.net/1/messages.json \\\n --data-urlencode \"token=YOURAPPTOKEN\" \\\n --data-urlencode \"user=YOURUSERKEY\" \\\n --data-urlencode \"title=Backend down\" \\\n --data-urlencode \"message=vigil API failed its liveness check\" \\\n --data-urlencode \"priority=1\"\n`\n\nOr the same thing from JavaScript, say inside a Cloudflare Worker or a Node service that just caught an error:\n\n`javascript\nawait fetch(\"https://api.pushover.net/1/messages.json\", {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/x-www-form-urlencoded\" },\n body: new URLSearchParams({\n token: env.PUSHOVERAPPTOKEN,\n user: env.PUSHOVERUSERKEY,\n title: \"5xx spike\",\n message: \"api.tigzig.com returning 500s\",\n priority: \"1\",\n }),\n});\n`\n\nThat single call is the whole thing. Everything else, which alerts fire, how loud they are, deduplication, recovery, is just your own logic around when to make the call.\n\nThe priority dial\n\nThe one feature I lean on most is the priority dial, which runs from -2 to 2 and decides how loud each alert is:\npriority -2, silent. No sound, no vibration, a pure log entry on the phone.\npriority -1, quiet. Shows up, no sound, look when you look.\npriority 0, normal. A normal buzz, respects your quiet hours.\npriority 1, high. Buzzes through quiet hours, even at night, look soon.\npriority 2, emergency. This is the one that keeps ringing. It re-alerts on a repeat interval and does not stop until you physically open the app and acknowledge it. If you sleep through the first buzz it rings again. It needs two extra parameters, retry (how often to re-ring, minimum 30 seconds) and expire (give up after, maximum 10,800 seconds, three hours, and Pushover caps it at 50 retries either way). I run retry 60 and expire 3600, so an unacknowledged emergency rings once a minute for up to an hour.\n\nYou can also set a different sound per alert type, so after a while you know the cert one from the CPU one by ear without looking. I am still on default sounds, that is a knob I have not turned yet.\n\nWhere I actually use it\n\nI run more than 50 apps on tigzig.com across two servers, and Pushover is wired across the whole fleet. Here are some of the things it watches, each one a small script that makes that one POST when it detects its thing, grouped by how loud I made each.\n\nEmergency, priority 2, rings until acknowledged, the get-out-of-bed tier:\nHost CPU alarm, the box's CPU pegged or a load spike.\n4xx velocity auto-block, a single IP hammering the API fast enough to auto-jail it.\nPublic API liveness outage, a backend failing its synthetic health probe.\nPublic API 5xx alarm, real callers getting 500s in the logs.\n\nHigh, priority 1, buzzes through quiet hours, the look-soon tier:\nEdge velocity jail, an IP got auto-jailed at the Cloudflare edge. The alert carries a deep link straight to the jail dashboard, since Pushover lets you attach a URL and a button to a message.\nOperator-range anomaly, activity from an IP outside my tight expected range, the \"is that actually me\" login check.\n\nQuiet, priority -1, the FYI tier:\nNew user feedback, someone left feedback in an app. I want to know, it is not a fire.\n\nNormal, priority 0, the all-clear tier:\nRecovery pings, when a firing alert clears I get a quiet RECOVERED message with the outage duration, so I know it is back without going to check.\n\nIt also quietly backs the TLS cert expiry warnings (warn under 21 days, critical under 7, because an expired origin cert silently takes a site down) and the public status-page rollup. The value here is that one simple channel, but by picking a priority per event it becomes a proper tiered pager, some events barely make a sound and the serious ones keep ringing until you answer.\n\nThe one pattern worth adding on top\n\nA naive setup fires the same alert every time the cron runs, so a three hour outage becomes ninety identical buzzes. I added a small active_alerts` table. The first time a problem is seen I insert a row and fire the Pushover, on every run after that the insert is a no-op because a unique key blocks the duplicate, so it stays silent. When the problem clears I delete the row and send one quiet RECOVERED with the outage duration. Net effect, one alert when it breaks, one when it heals, nothing in between. That fire-once-and-recover pattern is the single most useful thing to add on top of raw Pushover, and it is about fifteen lines of glue.\n\nThe brain behind it: the monitoring dashboard\n\nA related note, Pushover is only one part of the monitoring story .. it's just the notification arm. The brain behind it is a monitoring dashboard I built for myself and for my clients, the Tigzig Command Center.\n\nThere are great off-the-shelf tools for this (Grafana, Prometheus), but I wanted something exactly to how I think. It started last year as a tiny logging service and a single page, and grew over many months into what I run now.\n\nIt pulls everything into one view: live edge traffic, CF worker logs, down to the individual IP, which IPs and which /24 subnets are getting blocked, errors as they happen, the kinds of attacks landing through the day, TLS certs about to expire, and a lot more, tab by tab.\n\nThe eye-opener from watching at the IP level and raw hit level: the large majority of what hits a public endpoint is bots, scanners, crawlers, AI agents, and outright attackers, not real users. Actual human raw hits, in the territory of 1-3%.\n\nSelf-hosting, and the alternatives\n\nA common question is whether you can self-host it. Pushover itself, no, it is a hosted service with its own phone apps and its own connection to Apple and Google's push networks, you cannot run the Pushover server. If self-hosting is a hard requirement the two well known open source options are ntfy (ntfy.sh), which you can self-host and which has its own phone apps, and Gotify, a self-hosted server with its own Android app.\n\nOne point worth knowing, and it runs against the obvious instinct. For the specific job of \"tell me my server is down\", a hosted notifier like Pushover is the safer choice. If you self-host your notifier on the same infrastructure you are monitoring, then when the box or the network dies your notifier dies with it, and the one message you most needed, \"I am down\", never gets sent. A hosted independent service has no shared fate with your servers, so it can still deliver the bad news while your own stuff is on fire. Self-host the notifier only if you have a compliance reason to keep the message content in house, and even then, run it somewhere independent of the thing it watches.\n\nOn credibility, Pushover is a small, long-lived, boring-in-a-good-way company. It turned ten in 2022 and has been going since 2012. One-time purchase, no subscription for basic phone use, and a generous allowance of 10,000 messages a month for free, shared across all the applications on your account. For a solo dev or a small business it is one of those rare tools that is cheap, does one thing, and has not let me down. Bigger teams reach for heavier pager platforms like PagerDuty or Opsgenie, with on-call schedules and escalation policies, Pushover is the right-sized tool below that tier.\n\nI almost built this myself\n\nA bit of story here. I actually started building my own version of this with Claude Code before I looked around. I'd assumed this kind of thing might be enterprise level .. or require configuring some complex platform.\n\nExplored with Claude Code only ... Pushover came top of list and building my own stopped making sense. Top reasons:\nThe economics. A one-time $5 for 10,000 messages a month. To build and run my own? My hourly rate is many multiples, am swamped and turning back potential clients. Easy no.\nThe time. The build is simple, sure. But then you test it, you babysit it, something crashes at the wrong moment. No desire to take that on for an already-solved problem.\nThe infrastructure. I already run two servers that are critical to me, plus another two dedicated to clients, and just keeping those monitored and secure is a real job. A third box to host my own notifier (Oracle free tier, maybe) is one more thing to harden. Go serverless instead and that's its own headache.\n\nPushover is zero headache.\n\nThe idea is this: just because you can build it doesn't mean you should. As a one-man show it's an easy call. For a big team the math might be different.\n\nSend warnings, keep the secrets out\n\nOne rule I would keep whatever else you do. Every message passes through Pushover's servers and lands on your phone's lock screen, so the message should say something needs your attention, go look, and it should never carry the sensitive thing itself. No API keys, no passwords, no customer data in the body. My alerts are pointers, \"5xx spike on X\", \"cert expiring\", \"IP jailed, open the dashboard\", and where I want the detail I attach a link to my own authenticated dashboard rather than putting it in the notification. A lost or stolen phone then leaks a doorbell rather than a filing cabinet.\n\nYou can just ask your AI to wire it\n\nYou do not really need to study any of this. Pushover is standard enough that any current AI coding assistant already knows it cold. Tell your agent to add a Pushover alert when this backend fails its health check and it will drop in the exact POST above, wire it to your env vars, and you are done. I did not have to read a new doc to set mine up, the agent already knew how, the same way it knows how to send an email. If you want to go deeper the official docs are genuinely good and linked below, but the barrier to entry is close to zero.\n\nThe bigger capabilities I do not use\n\nFor completeness, here is the ceiling. I use the simple ninety percent case, one recipient, the priority dial, a deep link, dedup and recovery. The rest of the surface, in case you need it:\nMultiple devices and device targeting, register phone, tablet and desktop and send an alert to one specific device.\nA desktop and browser client, so alerts hit your laptop too.\nDelivery groups and Pushover for Teams, fan one alert out to a whole team with per-user routing, the step toward on-call without a full pager platform.\nInactivity monitors, a Pushover for Teams feature that pages you when an expected check-in goes silent, a dead man's switch for the job that was supposed to run and did not. I do not have this on my plan and it is the one piece I wish I did, since today I run a separate heartbeat monitor to cover the same gap.\nEmergency receipts and acknowledgement callbacks, for priority 2 alerts you can query who acknowledged and when, and have Pushover POST back to your own server the moment it is acknowledged. That is the classic inbound-webhook direction, which I do not currently use.\nWebhook receivers, a Pushover-hosted URL you point a system at when it can only POST arbitrary JSON and does not speak the Messages API, and Pushover turns that JSON into a notification. Added in early 2026.\nCustom and uploaded sounds, a distinct tone per alert class.\nImage attachments, attach a screenshot or a graph to the notification, up to 5 MB.\nMessage formatting, HTML or monospace and extra clickable URLs in the body.\nThe Glances API, push tiny data points to a phone or watch widget, a lightweight always-on readout separate from notifications.\nThe Subscriptions API, let other people subscribe to your application's notifications, which turns Pushover into a light broadcast channel.\nMessage TTL, have a notification auto-delete after N seconds so stale alerts clean themselves up.\n\nRoughly, I use it as a personal pager. It can also be a team pager, a broadcast channel, a desktop notifier, and a watch-face data feed.\n\nCost, once more\n\nIt is free for a 30 day trial, after which it is not free, a one-time $4.99 per platform covers it with no subscription on top. You get 10,000 free messages a month across your apps, and the company has been going since 2012. For a small builder it is about the best money-to-value ratio in the whole ops toolbox.\n\nLinks, for those who want the manual\nMain site: https://pushover.net\nThe messages API, the one POST: https://pushover.net/api\nEmergency priority and receipts: https://pushover.net/api#priority\nGlances API: https://pushover.net/api/glances\nSubscriptions API: https://pushover.net/api/subscriptions\nSelf-host alternatives: ntfy (https://ntfy.sh), Gotify (https://gotify.net)\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Security checklist, API + MCP hub, Browse all apps\n\nMore posts: Tigzig APIs - Reading the Logs, Shipping the Fixes. Status Page Live, R2 Downloads, Self-Correcting 4xxs., Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users, My Public MCP Server Got Hammered - Security Lessons from a Bot Attack, One Line of User Feedback, Fixed Within a Day. MFPRO Search Upgraded + Complete AMFI NAV History, 37M Rows, 2006 Onward."
},
{
"slug": "python-in-excel-claude-vs-xlwings-lite",
"title": "Python In Excel - Claude Vs. xlwings Lite? Who Wins?",
"tags": [
"claude-in-excel",
"python-in-excel",
"xlwings-lite"
],
"content": "Python In Excel - Claude Vs. xlwings Lite? Who Wins?\n\nPython In Excel - Claude Vs. xlwings Lite\n\nBoth can do Python in Excel so which is better?\n\nIt's like asking - Virat Kohli (Top Batsman) vs Lionel Messi (Top Footballer) - who's better with a ball? Just because both run around fields hitting a ball does not mean both are playing the same game\n\nxlwings, Claude in Excel, Jupyter/Colab (with AI) , Claude Code (VS Code) - all can do Python, Excel and ML Models. Each plays a different game - each a master of its game\n\nWould you put a Kohli to face Messi's penalty kick or Messi to bowl to Kohli. Imagine that ...right?\n\nThe question is - do you know which game you are playing?\n\nClaude in Excel\n= AI Assistant for Excel. Min $20 Pro subscription (includes Claude Code)\nFull Excel capabilities + Python sandbox + web search + local file upload\n\nBut can it do an API call? No.\nCan it do a web scrape? No.\nCan it connect to external databases? No.\nAutomation? No.\nProcess file directly on a local drive? No.\nCan you be sure it will give same result each time? No.\n\nIts job is different. Solid for Excel heavy lifting. Released recently, Used only for a few hours. It's powerful. And is going to change the way I work with Excel forever. More posts coming ...\n\nxlwings Lite\n= Pure python in Excel. Free\nAPI Calls, web scrapes, remote databases, local folders, PDF reports, HTML reports, Image processing ...\n\nIs it an Excel Assistant. No?\nCan it work autonomously in Excel? No (yet)\n\nIt's job is different - advanced automation, tools, full stack Excel apps.\nSee in action here: https://www.tigzig.com/xlwings-starter. 7 live apps, 3 hands-on guides and 17 reports. Battle scarred. Shared over the past one year\n\nWhat about ML Models\nFirst choice - Jupyter or Colab (AI Assisted). If model is nail then these are the hammers\nxlwings Lite - can do it all - from an xgboost to a neural net - but I find it a tad inconvenient for this\nClaude in Excel - fully autonomous - great choice with smaller / cleaner data. But there are trade-off. See here - https://www.tigzig.com/post/claude-in-excel\n\nWhat about FastAPI python backend?\nClaude Code (earlier Cursor)\nThe backends for all 30+ open source apps on tigzig.com are built this way. All Python scripts\n\nData processing pipelines?\nClaude Code\nMy IMDb Movies app has 230M rows sitting on a 16GB DuckDB database with a dual backend - one on Hetzner and another on an Oracle VPS. All Python scripts\nLive app: https://www.tigzig.com/movie-explorer\n\nReact /PHP / Windows desktop apps?\nClaude Code\nAll Tigzig apps built this way\n\nData Privacy?\nClaude in Excel - data goes to Anthropic servers.\nColab - data goes to Google servers\nClaude Code / Jupyter / xlwings Lite - all local\nBut...but...but ... there are important nuances - see here: https://www.tigzig.com/post/claude-in-excel\n\nCode Control\nClaude in Excel - no direct control - it runs on its own sandbox based on instructions. Though you can ask it to print the code it used. Full code control with all the rest\n\nFigure out the game you are playing - and then pick the tool\n\n\nRelated\n\nTools: Practice Lab, Technical Analysis in Excel, Databases + ML Models\n\nExplore: Claude for Analytics hub, Browse all apps, Guides index\n\nMore posts: Claude in Excel just one-shotted an XGBoost response model with train-test split, AUC and full decile table. In a spreadsheet., Claude in Excel & PowerPoint. Is it worth it? What works and what doesn't, Claude in Excel + MCP + xlwings Lite + Claude Code: Pull Data, Build a Model, Validate It., xlwings Lite Practice Lab - a free, hands-on guide for Excel professionals"
},
{
"slug": "python-in-excel-field-guide-practice-lab-for-ai-assisted-xlwings-lite",
"title": "Python in Excel: Field Guide & Practice Lab for AI-assisted xlwings Lite.",
"tags": [
"xlwings-lite",
"python-in-excel",
"ai-coders"
],
"content": "Python in Excel: Field Guide & Practice Lab for AI-assisted xlwings Lite\n\nI am sharing my workflow and documents for using AI to reliably generate xlwings Lite Python code. This is not a theoretical guide. It's a set of practical tools and hard rules which I use day to day in my client projects.\n\nWhat's in the xlwings Lite Practice Lab:\nAI Coder Instruction File: My personal 1,855-line directive file. This is the core context you feed the AI to ensure it get clean, reliable xlwings Lite code.\nStep-by-Step Field Guides: Three hands-on modules (Data Manipulation, Cleaning, Campaign Build) with the exact workbooks, guides and codes I use to test and validate.\nLive Apps: xlwings Lite apps: web scrapers, database connect, AI schema detection, machine learning, quants reports\n\nThe Core Protocol for AI code generation\nShow - Context is King: Give the AI everything - images, schemas, rules.\nTell - Instruct, Don't Assume: Be direct and iterate. One instruction is never enough.\nInspect - Trust, but Verify: All AI output must be validated. It makes mistakes.\n\nHow to get the materials\n\nGo to www.tigzig.com (Path: xlwings Lite Practice Lab)\n\nAll guides, workbooks, apps and the AI instruction file are available for download.\n\nxlwings Lite\n\nPart of my primary stack.\n\nCreated by Felix Zumstein.\n\nTIGZIG Co-Analyst\n\nThis is part of my open-source suite of 25+ tools for analysts and builders.\n\nxlwings Lite official site: lite.xlwings.org\n\n\nRelated\n\nTools: Practice Lab, AI Schema Detection, Databases + ML Models\n\nExplore: AI in Excel Hub, Quant Apps, All Guides\n\nMore posts: xlwings Lite Practice Lab - a free, hands-on guide for Excel professionals, The xlwings Lite AI Coder Instruction File - December 2025 Release, Which AI Coder should you use for xlwings Lite (Python in Excel)?"
},
{
"slug": "python-in-excel-with-claude-what-works-and-what-doesnt",
"title": "Claude in Excel & PowerPoint. Is it worth it? What works and what doesn't",
"tags": [
"claude-in-excel",
"python-in-excel"
],
"content": "Claude in Excel & PowerPoint. Is it worth it? What works and what doesn't. Plus a Python perspective\n\nClaude in Excel Evaluation\n\nI came to evaluate its Python-in-Excel capabilities. But what I started to use more is its Excel & PowerPoint (PP) heavy lifting. That is where the biggest value is. The Python sandbox is a bonus. The heavy lift for the attached deck was done by Claudes sitting in Excel and PowerPoint saving me many hours of manual labor and much pain.\n\nFast points\nIt pretty much does all your Excel work - analysis, pivots, formulas, stats, cuts, distros, scenarios etc etc ... does it fast. With nicely formatted outputs by default. Same with PowerPoint\nReads images + Web search - formatted tables and charts.\nBonus: you get Python power of analysis, charts, stats & ML.\nPython available in PowerPoint also. Useful for direct Python - PP\nBiggest value is in its 'human like' and intelligent Excel working, allowing you to focus on results (and validations).\nIs it worth it? Yes. Starts at $20. Plus you get Claude Code bundled. Your Excel and PowerPoint productivity impact is in multiples not percentages.\n\nTop Tip : Use voice dictation. 3x faster than typing - from Boris Cherny, creator of Claude Code (link below for his top tips). Win+H on windows & Fn Twice on Mac.\n\nFor more details see attached document and my previous posts. Some key highlights below.\n\nPython side useful stuff\n\nUpload CSVs into Pandas, run statistical tests, forecasts, regressions, ML models (work well for smaller/ cleaner datasets), advanced charts. Upload and process Excel files, PPTs, PDFs, JSONs.\n\nPython side Limitations\n\nSandbox is isolated - no API calls, no external databases, can't write to local drive. Files created inside the sandbox cannot come out (except images - see doc). 30MB upload limit. Many libraries installed but not all fully usable. Non-deterministic - same prompt may give different results. No automation possible. Data goes to Anthropic servers (there are nuances - see previous posts below)\n\nHow does this compare with xlwings Lite?\n\nI covered that in detail in my previous posts (links below). They are different tools for different jobs.\nxlwings Lite: full Python environment sitting inside Excel.\nClaude in Excel: Intelligent AI assistant with very useful Python powers.\nxlwings Lite: Do automations, tools and full stack Excel apps.\nClaude: Do all your Excel work\nxlwings Lite: Python power.\nClaude: AI power\n\nKnow the trade-offs, know the limitations, and pick the right tool.\n\nUseful Posts\nClaude in Excel just one-shotted an XGBoost response model\nPython In Excel - Claude Vs. xlwings Lite? Who Wins?\n2026 Infra Guide for AI Tool Builders - Part 1: AI Coder\nSee xlwings Lite in Action - Live Apps & guides\nClaude Code: Top 10 Tips from Boris Cherny (Verbatim)\n\n\nRelated\n\nTools: Practice Lab, Technical Analysis in Excel, QRep - Security Reports\n\nExplore: Claude for Analytics hub, Quant Apps, Browse all apps\n\nMore posts: Python In Excel - Claude Vs. xlwings Lite? Who Wins?, Power User Guide to Claude in Excel & PowerPoint - 26 Working Tips, Claude in Excel just one-shotted an XGBoost response model with train-test split, AUC and full decile table. In a spreadsheet., MCP Guide for Claude in Excel - and any MCP client"
},
{
"slug": "python-in-excel-with-xlwings-lite-part-2-connect-to-remote-databases",
"title": "xlwings lite |Connect to Remote Databases",
"tags": [
"xlwings-lite",
"python-in-excel",
"database-ai"
],
"content": "xlwings lite |Connect to Remote Databases\n\nAutomate, pull, analyze, report, chart, model.\n\nxlwings DB Connection\n\nCheck out detailed info REX AI Co Analyst site\nFull templates\nHow-to guides\nBackend code\nTechnical breakdowns\nResource Links\n\nDetailed info , source codes, how-to guides, resources and templates available on REX\n\nPython in Excel: with a mini code editor sitting inside Excel, a robust Excel objects model and a simple setup, xlwings Lite has nailed it. One of my primary tools now - alongwith Colab, Cursor, and Jupyter+Mito AI.\n\nIn this Part 2 on xlwings Lite, I show another powerful feature - API calls…lets you connect to any external system with API access.\n\nIn this demo: connect to a remote Postgres database, explore tables, pull records, and run custom SQL - all from Excel, using a simple spreadsheet interface. Excellent for automated data pulls and DB updates too.\n\nOnce data is pulled, we run exploratory data analysis with extended descriptive stats, frequency tables, distribution plots, charts - followed by an XGBoost response model with evaluation sheet, decile table, ROC/Gains chart, and scoring.\n\nThis is demo with mock data, but shows how xlwings Lite handles:\nComplex Python workflows\nExternal system connections via API\nRemote database connections\nExcel object manipulation - sheets, rows, columns\nAdvanced automation\nExcel-based tools with Excel as the UI\nCustom tables, reports, and charts\nStatistical analysis\nMachine learning model builds\n\nRemote DB Connections\n\nRequires a web layer - easy to manage. I use a custom built FastAPI server (open source), which you're free to use (just note: your credentials would pass through my server) - or you can deploy your own. Takes just a few minutes on Render. It's a reusable component I use to connect any frontend or UI to backend databases. I've covered this in earlier posts and videos - links in comments\n\nAPI Calls to LLMs\n\nNot covered in this video, but I've also tested API calls to OpenAI, Gemini, and Claude Sonnet using JSON schemas for structured output. Big deal for automation. Planning to cover this in a future part. But it's essentially the same - straightforward API call, parse the response, use it in your workflow.\n\nInstructions for AI Coders / LLMs\n\nFor humans too. In my Github Repo.\n\nxlwings Lite\n\nFrom Felix Zumstein, creator of the original xlwings - a solid, established Python package for Excel automation. He's laid out a strong roadmap and shared excellent documentation covering usage, API reference, limitations, and comparisons with other Python-in-Excel tools.\n\nhttps://lite.xlwings.org\n\nVideo\n\nFull 20 minute video on YouTube with timestamps: Demo, Caveats, FastAPI setup & Code Walkthrough\n\nhttps://youtu.be/6Vmzd4sHcxY\n\nAI Co-Analyst: explore 15+ open-source AI analytics apps at tigzig.com - AI Advanced Analysis App: https://lnkd.in/g8xPQvb8\n\nAI Automation Micro App - Fund Disclosures Excel to CSV https://lnkd.in/gxVbCdpv\n\nGitHub repo with demo files and scripts for xlwings Lite\n\nhttps://github.com/amararun/shared-xlwings-docs\n\nFastAPI Server - Latest Code\n\nhttps://github.com/amararun/shared-fastapi-rex-db-coolify\n\nFastAPI Server - How to Deploy on Render\n\nhttps://lnkd.in/g2A9h8f2\n\nThis guide shows how to connect ChatGPT with Supabase. The Render deployment steps are the same - just use the latest code (above). No environment variables needed. For live use cases, API keys and additional security advised.\n\nGet free Postgres / MySQL database quickly\nneon.tech - Postgres database, instantly available. 500MB free.\nSupabase - Postgres with auth, API, storage. 500MB free tier.\nAiven - Postgres and MySQL. 5GB free tier.\n\nxlwings Lite official site: lite.xlwings.org\n\n\nRelated\n\nTools: Databases + ML Models, Practice Lab, Database AI Apps\n\nExplore: AI in Excel hub, API/MCP directory, Guides hub\n\nMore posts: AI + Python in Excel with xlwings Lite - LLM API Calls | Part 3, Talk to Your Database from Excel - Postgres, DuckDB - via Claude in Excel with MCP, Connect Excel to a Database with Tens of Millions of Rows? Tigzig.com Has 15+ Apps and Guides for That Alone. Just Ask Your AI."
},
{
"slug": "python-in-excel-xlwings-lite-with-natural-language-instructions",
"title": "Python in Excel (xlwings Lite) with Natural Language Instructions.",
"tags": [
"xlwings-lite",
"python-in-excel"
],
"content": "Python in Excel (xlwings Lite) with Natural Language Instructions\n\nDictate logic → AI codes. My Workflow for AI Code Generation.\n\nWhat I cover\nxlwings Lite - quick start\nRobust workflow - top 5 Rules\n\nResults matter - code is incidental\nWhat doesn't matter: The perfect python code\nWhat matters: Delivering incremental impact without fatal errors.\n\nWhat does not work\n\nSome one-shot, in-the-cloud prompt. That's demo theatre. Fine for concept demo - they don't work on live projects.\n\nWhat works - My 5 hard rules\nBe specific: High level gyan does not work\nIterate: One step at a time. Validate. Debug. It is tedious. At times frustrating.\nReview: Demand plan + pseudocode from AI. Scan code if you can.\nValidate: Sacrosanct. An engineer doesn't 'quick scan' a plane before takeoff. Same principle.\nAI Audit: Final pass. Feed AI complete code in new session. It will still catch misses.\n\nTools - AI Coders\n\nGemini 2.5 Pro → my top pick. Huge free tier. 1M context. aistudio.google.com\n\nChatGPT → context fills fast + 2.5 Pro better.\n\nCursor + Claude Sonnet 4 → excellent. But they have tightened usage limits. I keep it for app builds.\n\nTools - Xlwings Lite AI Coder Instruction File\n\nMy personal 1855-line AI Coder instruction set for xlwings Lite. I use it on every project. Shared on app site. Optional, but makes code gen efficient\n\nDictate - don't type\nWhat does not matter: your typing speed\nWhat matters: a robust workflow, expressing your requirements and business logic to AI, and validating results\nVoice - It's fast and frictionless. Use Windows preinstalled tool (Win+H) and Mac (Fn key twice)\n\nAre you a Karpathy?\n\nMy first line of code in 1994. Last major 'manual coding' in 2022. I've executed projects with ML models, campaigns, and automations on Python/SAS with millions of dollars in tracked, signed-off impact. Now → AI-assisted. From automations to ML models. Faster. Better. Greater Impact\n\nUnless you're an Andrej Karpathy or Przemyslaw Debiak, let AI Code for you\n\nxlwings Lite\nPart of my primary stack.\nBeyond automations, build standalone apps connected to databases and custom backends\nCreated by Felix Zumstein.\n\nMaterials\n\nGo to www.tigzig.com → xlwings Lite Practice Lab.\n\nGuides, workbooks, apps, instruction file - all there.\n\nxlwings Lite official site: lite.xlwings.org\n\n\nRelated\n\nTools: Technical Analysis in Excel, Data Importer, AI Schema Detection\n\nExplore: xlwings Lite Practice Lab, AI for Excel hub, Guides hub\n\nMore posts: Which AI Coder should you use for xlwings Lite (Python in Excel)?, The xlwings Lite AI Coder Instruction File - December 2025 Release, Python in Excel: Field Guide & Practice Lab for AI-assisted xlwings Lite."
},
{
"slug": "python-workflows-inside-excel-with-xlwings-lite-free",
"title": "Python Workflows. Inside Excel. With xlwings Lite (free) - Powerful.",
"tags": [
"xlwings-lite",
"python-in-excel"
],
"content": "Python Workflows. Inside Excel. With xlwings Lite (free) - Powerful.\n\nData transforms, automations, custom functions, API Calls, stats, ML\n\nAnother superb addition to analyst and data scientist's toolkit - xlwings Lite. From Felix Zumstein, creator of xlwings - the powerful and widely used Python package for Excel automation. I've used xlwings off and on over the years - solid package.\n\nxlwings Lite\n\nCheck out dedicated xlwings LIte section on REX AI Co Analyst site\nFull templates\nHow-to guides\nBackend code\nTechnical breakdowns\nResource Links\n\nxlwings Lite was released just yesterday. Ran a few quick codes- what I typically run in Jupyter / Colab - it's brilliant.\n\nSetup was simple - just add the xlwings Add-in and you're set. Tried an XGBoost model on cleaned-up mock data - decile table, scoring, ROC/Gains chart ...just to see how it handles...Also some quick analysis, feature engineering, basic stats - smooth and easy.\n\nMost impressive\nManipulate Excel objects - sheets, columns, rows\nTransfer data between Excel and Python\nNeat mini editor\nConsole for print() output and debugging\nRequirements.txt for package installs\nCustom Functions\nWeb API calls (huge)+ DB access via same layer (yet to try out)\n\nQuick Tips (based on initial hands-on)\nFormat Excel ranges as Tables - good practice anyway.\nUse any LLM or AI coder - they already \"know\" the xlwings object model.\nWrite everything as standard Python functions (def).\nUse @func to create Excel-callable functions (=formula), and @script for direct code runs.\nGreat docs and API reference if you get stuck.\n\nIs it a replacement for Jupyter/Colab?\n\nNopes..many analytics and ML workflows are iterative and need data warehouse connects. A strong addition to the toolkit, not a replacement.\n\nGreat fit for\nExcel automation - I'll take Python over VBA any day (and the combo - even more powerful)\nComplex data transformations - no need to switch between Jupyter and Excel.\nIn-Excel EDA, validations, diagnostics\n\nLinks\n\nExplore 15+ open-source AI analytics apps at tigzig.com with natural language to SQL, python charts, stats, database connectivity and more.\n\nFree access, source codes and build guides\nAI Powered Automation - Connect ChatGPT with n8n\n\nxlwings Lite official site: lite.xlwings.org\n\n\nRelated\n\nTools: Practice Lab, Technical Analysis in Excel, Databases + ML Models\n\nExplore: AI in Excel hub, Browse all apps, APIs and MCP\n\nMore posts: xlwings Lite Practice Lab - a free, hands-on guide for Excel professionals, Python In Excel - Claude Vs. xlwings Lite? Who Wins?, Which AI Coder should you use for xlwings Lite (Python in Excel)?"
},
{
"slug": "qdesk-one-app-three-ways-in-sep2026",
"title": "Everything I Built for Market Reports Now Sits in One App. QDESK, and Three Ways In.",
"tags": [
"portfolio-quants",
"mcp-servers"
],
"content": "Everything I Built for Market Reports Now Sits in One App. QDESK, and Three Ways In.\n\nEverything I had built for market reports now sits in one app. It used to be three separate ones, so people had to work out which was which, and I had to make the same change in three places.\n\nIt is called QDESK, and there are three ways in. You can ask the agent in plain English. You can fill a form and hit generate. Or you can point your own code, or your own agent, at the API and MCP, which is what the first two are calling underneath.\n\nWhich one suits you depends on what you are working with. A chat window cannot install anything, so it comes in over MCP. Something like Claude in Excel will install what it needs and do the work itself.\n\nOne change came out of reading the logs. The agent now looks up the symbol live. It used to recall tickers from memory, and that guessing was where people were getting stuck.\n\nFree, no signup.\n\nThe reports are the same whichever way you come in\n\nSecurity performance, AI technical analysis on the daily and weekly charts, a QPulse tearsheet against a benchmark, QREP with around eighty risk and return figures for up to six securities side by side, company financials, price data, and symbol search. Indian mutual fund NAV history comes from AMFI.\n\nIf you would rather not start from a blank box, there are sample prompts on the page. One of them runs the whole set of reports in one go, so give that one a couple of minutes. The single reports come back quicker.\n\nIndian mutual fund NAV history\n\nThe agent will also pull Indian mutual fund NAV history and hand it over to you.\n\nIf you want to pick the funds yourself and get a CSV, that is in MFPRO, on the Data and API tab. Single fund or in batches of 20. Every scheme from April 2006, updated three times a day. Free, no sign in.\n\nWhere to go\n\nFor you: tigzig.com/qdesk\n\nFor your agent: api.tigzig.com\n\nFor the API docs: tigzig.com/apis. Those are the same endpoints the agent is calling.\n\nFor the fund NAV picker: tigzig.com ➜ MFPRO ➜ Data and API\n\nQDESK: quant and portfolio reports in one app, with three ways in - ask the agent, fill a form, or point your own code or agent at the API and MCP"
},
{
"slug": "qdesk-reporting-desk-sep2026",
"title": "QDESK: A Reporting Desk for Portfolio Reports and Financial Data. Just Ask in Plain Language.",
"tags": [
"portfolio-quants",
"mcp"
],
"content": "QDESK: A Reporting Desk for Portfolio Reports and Financial Data. Just Ask in Plain Language.\n\nQDESK is a reporting desk to generate portfolio reports and pull financial data. Just ask in plain language. An agent is sitting there to run the reports and guide you.\n\nData is from Yahoo Finance.\n\nYou do not need to know the ticker. Give it the name of the company, crypto, commodity or index, and it will look up the symbol first and then run the report.\n\nWhat it produces\n\nA security performance report, technical analysis on daily and weekly, a QPulse tearsheet against a benchmark, and QREP metrics with around eighty risk and return figures for up to six securities side by side. It also pulls company financials, prices, and NAV history for Indian mutual funds.\n\nThere are sample prompts, so try them out. One of them runs the whole set of reports in one go.\n\nQDESK: tigzig.com/qdesk\n\nAPI and MCP\n\nIt is also available on API and MCP if you would like to connect it up to your own tooling. I see many users pulling data directly into their Power Query, Excel and Google Sheets.\n\nEach report is its own service, so you can call just the one you want rather than the whole set. Security performance, technical analysis, QPulse and QREP are separate endpoints, and so are the Yahoo Finance data and the mutual fund NAV history.\n\nAll of them are available as plain REST APIs, and most are on MCP as well, so you can pull them into your own tooling or hand them to an agent. The hub says which.\n\nFor Excel and Sheets, the Yahoo Finance endpoints have sheet-ready variants of the statements, so the rows come back in a shape you can drop into a worksheet without reshaping them first.\n\nThe hub has all of them, with what each endpoint takes and what it returns: tigzig.com/apis\n\nQDESK: a reporting desk for portfolio reports and financial data, with the reports it produces and the API and MCP endpoints behind them"
},
{
"slug": "qrep-claude-in-excel-mcp-jun2026",
"title": "QREP in Excel: 81 Risk and Return Metrics Inside Your Sheet Through Claude in Excel.",
"tags": [
"mcp",
"claude-in-excel",
"portfolio-analytics",
"excel"
],
"content": "QREP in Excel: 81 Risk and Return Metrics Inside Your Sheet Through Claude in Excel.\n\nPublished: 2026-06-29\n\nStandalone one-pager release: QREP, a Tigzig MCP server, brings 81 risk and return metrics straight into Excel through Claude in Excel - asset returns, drawdowns, Sharpe, Sortino, alpha, beta, volatility and more. Just ask Claude in Excel things like 'Compare gold and bitcoin, last 10 years' or 'compare Mag 7 past 10 years' - any stock, commodity or crypto that's on Yahoo Finance. Change symbols, time period, run cuts your own way, format it your way, push it into a deck, or do it directly in PowerPoint. MCP guide at tigzig.com -> MCP -> QREP. For AI-coders, just ask Gemini / ChatGPT / Claude to help connect Claude in Excel to QRep from tigzig.com - it walks through. Wary of someone else's MCP server? The code is open source, a single Python file - ask your AI to deploy the QREP MCP server from tigzig.com. Free to use. Powered by Ran Aroussi's QuantStats. Three companion guides linked: Claude Cowork Guide, Claude in Excel & PowerPoint 26 working tips, and Claude as the universal agent.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/qrep-claude-in-excel-mcp-jun2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/qrep-claude-in-excel-mcp-jun2026 with header Accept: text/markdown\n\nTags\nmcp, claude-in-excel, portfolio-analytics, excel\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: QRep - Security Reports, Quants Agent\n\nExplore: Claude for Analytics, Python in Excel lab, APIs & MCP catalog\n\nMore posts: MCP Guide for Claude in Excel - and any MCP client, Connect Excel to a Database with Tens of Millions of Rows? Tigzig.com Has 15+ Apps and Guides for That Alone. Just Ask Your AI., Claude Cowork and Claude Code Have Quietly Become My Universal Front End. More Than Excel, Jupyter or Colab."
},
{
"slug": "qrep-excel-webservice-81-metrics-sep2026",
"title": "Pull 81 Risk and Performance Metrics Into Excel With One Formula",
"tags": [
"portfolio-quants",
"excel"
],
"content": "Pull 81 Risk and Performance Metrics Into Excel With One Formula\n\nDid you know you can pull 81 risk and performance metrics for a stock, crypto, commodity or FX pair straight into Excel with one formula? It is powered by the Tigzig API. Copy the formula below into a cell, hit enter and give it a few seconds.\n\nIt uses the Excel WEBSERVICE function to make an API call. That works well for many cases. It has some limitations too, and those are further down.\n\nThe formula\n\nPaste this into a single Excel cell and hit enter.\n\n``\n=TEXTSPLIT(WEBSERVICE(\"https://qrep-api.tigzig.com/v1/compare?symbols=MSFT,META,ORCL,GOOGL,AMZN&benchmark=SPY&startdate=2025-03-14&enddate=2026-09-14&riskfreerate=0.05&format=tsv\"),CHAR(9),CHAR(10),FALSE,0,\"\")\n`\n\nIf you get a warning, hit 'Paste without active content'. If it gives an error, go to the end of the formula and hit enter again.\n\nSee the raw data in a browser\n\nPaste this into your browser if you just want to look at the numbers.\n\n`\nhttps://qrep-api.tigzig.com/v1/compare?symbols=MSFT,META,ORCL,GOOGL,AMZN&benchmark=SPY&startdate=2025-03-14&enddate=2026-09-14&riskfreerate=0.05&format=tsv\n`\n\nYour browser will ask you to save a file named 'compare'. Save it, open it with Notepad and copy the values straight into Excel. Swap format=tsv for format=json` if you would rather read it on screen. You can get the same thing from the tool itself.\n\nOne caveat on WEBSERVICE\n\nEach call is one API request from one cell. That is fine for a single request like this one. Point it at a few hundred symbols and it stops working well. Excel sends all those calls out at once, so you run into the host's rate limit and get errors back for most of them. Power Query handles that case properly. So does Python inside Excel, which for me means xlwings Lite. I have written about both elsewhere.\n\nParameters\n\nThe risk free rate is a decimal fraction. 0.05 means 5 percent, and passing 5 is rejected. Six symbols is the maximum per call, and the benchmark sits outside that count. Dates are plain YYYY-MM-DD.\n\nThe tool\n\nThis is part of Tigzig's QDESK set of market tools. It is available via API, via MCP, through an AI agent interface, and through a form if you would like to do it yourself. It works for any Yahoo Finance ticker with data.\n\nFor you: tigzig.com/qrep\n\nFor your agent: api.tigzig.com\n\nCredit\n\nThe backend is powered by the Python QuantStats library, github.com/ranaroussi/quantstats. I have made some changes to the calculations. Those are covered in the Docs section of the tool.\n\nPull 81 risk and performance metrics into Excel with one formula"
},
{
"slug": "qrep-mcp-quants-numbers-claude-chatgpt-excel",
"title": "Pull your quants numbers directly with Claude / ChatGPT in Excel (with MCP).",
"tags": [
"mcp",
"quantstats",
"excel",
"open-source"
],
"content": "Pull your quants numbers directly with Claude / ChatGPT in Excel (with MCP).\n\nQREP Security Performance Report\n\nAsk in plain English: \"Pull all stats for Apple, Google, Microsoft past 10 years\".\n\nGet back 81 metrics - returns, sharpe, sortino, drawdowns - the whole bucket load.\n\nPowered by Ran Aroussi's (creator of yfinance python library) QuantStats library.\n\nWorks with Claude in Excel, ChatGPT in Excel, Claude Desktop, Cursor, n8n, or any MCP client.\n\nMCP Server URL: qrep-api.tigzig.com/mcp\n\nThe server is free to use (rate limits apply). Just add that URL as custom connector and you are set.\n\nThis is the public version. No API key. No auth. For production use, always use OAuth / SSO (see links below).\n\nAt the same time, my standard MCP advice still stands - don't connect to unknown MCP servers blind. Use servers from reputed providers, or ones you deploy yourself. If you'd like to implement, easiest way - paste this message to your AI Coder. It knows what to do.\n\nFull source code is open. Single Python file. Clone it, inspect it, deploy your own.\n\nWeb version of QREP: qrep.tigzig.com\n\nResources\nGitHub - QuantStats - The library powering the analysis\nGitHub - QREP MCP - Full source code, single Python file\nDocs - QREP MCP - API documentation\nMCP with OAuth - How to secure MCP servers for production use\n\n\nRelated\n\nTools: QRep - Security Reports, Quants Suite, Quants Agent\n\nExplore: Claude for Analytics hub, Quant tools hub, MCP servers hub\n\nMore posts: QREP in Excel: 81 Risk and Return Metrics Inside Your Sheet Through Claude in Excel., S&P 500 Drawdown Analysis with QREP, MCP Guide for Claude in Excel - and any MCP client, QRep - Powered by QuantStats. Live Now., Three Tools I Built for Market Analysis - QSUITE, TREMOR, QREP"
},
{
"slug": "qrep-quantstats-security-analytics-live",
"title": "QRep - Powered by QuantStats. Live Now.",
"tags": [
"portfolio-analytics",
"technical-analysis"
],
"content": "QRep - Powered by QuantStats. Live Now.\n\nQRep Security Analytics\n\nSecurity analytics tools with technical analysis module.\n\nAround 90+ KPIs across risk, returns, ratios and drawdown, HTML and PDF reports, CSV downloads...plus interactive technical analysis charts...and you can change the parameters on the fly. Search is instant - no need to go Google the Yahoo Finance symbol first.\n\nBuilt on top of Ran Aroussi's QuantStats (recently updated with major enhancements) and yfinance libraries.\n\nResources\nLive tool: https://tigzig.com/qrep\nSource code: In Docs on app\nPart of the Tigzig Analyst suite\n\n\nRelated\n\nTools: QRep - Security Reports, Quants Agent, Quants Suite\n\nExplore: Quant apps hub, MCP and API hub, Analysis hub\n\nMore posts: Three Tools I Built for Market Analysis - QSUITE, TREMOR, QREP, Pull your quants numbers directly with Claude / ChatGPT in Excel (with MCP)., S&P 500 Drawdown Analysis with QREP, QREP in Excel: 81 Risk and Return Metrics Inside Your Sheet Through Claude in Excel."
},
{
"slug": "qsuite-nifty-sp500-technical-analysis-llm-comparison",
"title": "Nifty and S&P 500 Down - Run a Technical Analysis Check. 9 LLMs Compared.",
"tags": [
"technical-analysis"
],
"content": "Nifty and S&P 500 Down - Run a Technical Analysis Check. 9 LLMs Compared.\n\nsrc: /images/blog/qsuite_02.png\nalt: QSUITE Technical Analysis Report\n\nNifty down ~7% from its January peak. S&P 500 down ~4%. Good time to run a technical analysis check on your holdings.\n\nMy QSUITE generates AI technical analysis reports - daily and weekly charts, support/resistance, MACD, RSI, Bollinger Bands, volume analysis, price outlook. PDF and HTML output.\n\nNew - you now get 9 LLM choices including GPT 5.4 and Claude Sonnet 4.6 (both just released). Default is Gemini 3 Flash. Premium models take longer but go deeper.\n\nI ran all 9 models on S&P 500 and Nifty 50 for the same time period - reports are here if you want to compare how different LLMs read the same charts.\nView all 9 model comparison reports\n\nI use this as a second opinion tool. Not financial advice.\n\nFree to use (with rate limits). No Login. No ads. Open Source.\n\nqsuite.tigzig.com\n\n\nRelated\n\nTools: Quants Agent, Technical Analysis, QRep - Security Reports\n\nExplore: 9 LLM comparison reports, Quant apps hub, API and MCP hub\n\nMore posts: Three Tools I Built for Market Analysis - QSUITE, TREMOR, QREP, Quants Agent: Now with LLM Choices for Technical Analysis Reports, Bitcoin Down nearly 30% in 25 days. What Does AI Technical Analysis Say?"
},
{
"slug": "quants-agent-llm-choices-technical-analysis-reports",
"title": "Quants Agent: Now with LLM Choices for Technical Analysis Reports",
"tags": [
"technical-analysis",
"portfolio-analytics"
],
"content": "Quants Agent: Now with LLM Choices for Technical Analysis Reports\n\nThe AI Technical Analysis report (support/resistance, indicators, price outlook, volume, daily/weekly charts) now lets you pick your LLM - including GPT-5.2 and Claude Sonnet 4.5.\n\nHow to use it\n\nGo to quants.tigzig.com\n\nAsk the agent: \"Technical analysis report for Nifty 50. What model choices do you offer?\"\n\nHow it works behind the scenes\nFetch: Daily & weekly OHLCV from Yahoo Finance (yfinance library)\nCalculate: EMAs, MACD, RSI, ROC, Bollinger Bands (finta library)\nChart: matplotlib generates price/volume/indicator charts\nAnalyze: LLM gets charts (base64) + data tables + structured prompt via OpenRouter\nUpload: LLM Analysis goes to my Markdown to PDF service\nConvert: Markdown to HTML to styled PDF\nOutput: Downloadable PDF + interactive HTML\n\nOpen Source\n\nHit 'Docs' for all source codes.\n\n\nRelated\n\nTools: Quants Agent, Technical Analysis, Technical Analysis in Excel\n\nExplore: Quant Apps hub, APIs and MCP, Build guides\n\nMore posts: [\\[ENHANCEMENT\\] AI Technical Analysis Now Supports Multiple LLM Choices](https://www.tigzig.com/post/enhancement-ai-technical-analysis-now-supports-multiple-llm-choices), Nifty and S&P 500 Down - Run a Technical Analysis Check. 9 LLMs Compared., Build AI Workflows with MCP Servers + n8n"
},
{
"slug": "quants-suite-5-reports-performance-risk-technical-analytics",
"title": "Quants Suite. 5 Reports. Performance, Risk & Technical Analytics.",
"tags": [
"portfolio-analytics",
"technical-analysis"
],
"content": "Quants Suite. 5 Reports. Performance, Risk & Technical Analytics.\n\n5 Reports. Performance, Risk & Technical Analytics across Stocks, Indices, Commodities and Crypto. PDF, HTML, CSV outputs. Open Source. Live.\nSecurity Performance Report\nMultiple securities in a single view\nCompare across asset classes - e.g. Google vs. S&P vs. Brent (F) vs. Crypto vs. Gold (F)\nMetrics & Charts: CAGR, Sharpe, Sortino, Drawdowns, Monthly Returns\nCustomize: Risk-free rate, performance period\nFormat: HTML Report + Interactive Chart\nAI Technical Analysis\nDual timeframes: Daily and Weekly\nIndicators: EMAs, Bollinger Bands, MACD, ROC, RSI\nAI Generated Analysis (Google Gemini 2.5 Flash)\nSections: S/R Levels, Indicator Analysis, Pattern Recognition, Volume Analysis & Technical Outlook with base case, confirmation and invalidation\nFormat: PDF & HTML\nQuantStats Report\nWrapper over QuantStats Python package\nSecurity vs. benchmark: performance, risk and drawdown metrics\n12+ Charts & 70+ Metrics\nFormat: HTML Report\nFinancials & Profile\nFinancials: P&L, B/S, Cash Flow, Quarterlies from Yahoo Finance\nSnapshot: Snapshot info from Yahoo Finance. 170+ data elements\nFormat: CSV & Excel Downloads\nPrice Data\nPrice data from Yahoo Finance\nCustomize the period\nFormat: CSV & Excel Download\n\nLive App\n\nquants-suite.tigzig.com\n\nOpen Source. Docs: On app site. Hit 'Docs'\n\n\nRelated\n\nTools: Quants Suite, QRep - Security Reports, Quants Agent\n\nExplore: Quant Apps hub, Security Performance Report API, Technical Analysis API\n\nMore posts: Three Tools I Built for Market Analysis - QSUITE, TREMOR, QREP, Security Performance Report for Investors. AI Quant Agent. Live. Open Source. Free., QRep - Powered by QuantStats. Live Now."
},
{
"slug": "quantstats-four-routes-qrep-agent-aug2026",
"title": "Comparing Securities on Returns, Risk and Drawdowns. Four Routes into QuantStats, Depending on What Your AI Can Do.",
"tags": [
"portfolio-analytics",
"mcp",
"ai-coders"
],
"content": "Comparing Securities on Returns, Risk and Drawdowns. Four Routes into QuantStats, Depending on What Your AI Can Do.\n\nIf you want to compare several securities on returns, risk, drawdowns and tail risk, QuantStats (open source) by Ran Aroussi (creator of yfinance) is a wonderful library. How you use it depends on what your AI can do and what you want to do.\n\nIn an AI chat like ChatGPT, it cannot install Python for you, so you connect it to an MCP server. My QREP MCP server is one option.. the backend is powered by QuantStats. The MCP server is also open source in case you want to deploy your own.\n\nWith an AI agent like Claude Cowork / Claude in Excel / Claude in PowerPoint / ChatGPT Work it installs the package itself, runs the numbers, and formats the output however you ask, or builds you a dashboard. It can also call the QREP API directly, which returns a fixed set of 81 metrics and skips the setup step. As well as connect to MCP. Whichever way you want to.\n\n(Sample prompt for the API option is in the section below.)\n\nAnd if you would rather run cuts interactively and look at them on screen, the QREP tool does that in the browser and also generates the QuantStats format tear sheet.\n\nThe tool and the API are both powered by QuantStats and both open source. A few implementation choices in QREP mean a handful of numbers differ slightly, written up in the methodology page.\n\nAnd here is the beautiful part if you are working with an AI Agent - since it has python, and if you are comfortable diving into formulas you can ask it to do the python functions from scratch. Does it pretty fast. Great for validations or even otherwise.. especially if you need metrics like rolling returns ..\nQuantStats GitHub Repo (Ran Aroussi): github.com/ranaroussi/quantstats\nQREP Tool: tigzig.com/qrep\nQREP API / MCP Docs: tigzig.com/apis/qrep\n\nOne clarification on Python inside a chat app\n\nRegarding python capability in a chat app like ChatGPT / Claude.ai, there are some nuances. Both of them have a python sandbox and ChatGPT has had that for a very long time, but the capabilities vary - on what is already installed, if they can install a new one, outside / API access etc. ChatGPT is fairly more restricted than Claude.ai chat. So e.g. Claude.ai can run yfinance and quantstats, ChatGPT chat cannot. However their cowork apps i.e Claude Cowork and ChatGPT Work (desktop apps) are a different animal ... and have far more flexibility including writing to your local drive.\n\nDeserves a separate post in itself, but in the meanwhile simplest is to ask it what capabilities it has and what it can / cannot do, and have it test it out for you .. try it out with the prompts below .. or with a test data file that you upload.\n\nFor Cowork capabilities, check out the hub below on TigZig: Claude for Analytics\n\nWhat to actually say to your agent, either route\n\nTo have it run QuantStats itself:\n\n``\ninstall quantstats and use it to analyze the Mag 7 against the S&P 500 over the\npast 5 years, pull out the metrics for returns, risk, alpha, beta, drawdowns\netc ...share results as a html dashboard\n`\n\nTo use the API:\n\n`\nhelp me analyze the Mag 7 against the S&P 500 over the past 5 years, use the\nQREP API from tigzig.com.. share result as a html dashboard\n``\n\nIf you are running the tool\nNo need to punch in yahoo finance.. search and select there only.\nUse the export function to get the whole metrics table as well as the raw data of prices as a CSV.\nAdvanced metrics parameters allows you to modify the omega threshold, var/cvar confidence & tail cutoff.\nThe tool also has technicals - EMAs, RSI, MACD etc. These are computed in the browser using the adjusted close prices from the backend. The docs section has the exact formulas used.\nRefer to docs for some implementation choices, reason for variances as well as source code to the frontend and backend (integrated API / MCP server).\n\n\nRelated\n\nTools: QRep - Security Reports, Quants Agent, TREMOR - Macro Stress Signals\n\nExplore: QREP API + MCP docs, Claude for Analytics hub, API + MCP hub\n\nMore posts: QREP in Excel: 81 Risk and Return Metrics Inside Your Sheet Through Claude in Excel., Pull your quants numbers directly with Claude / ChatGPT in Excel (with MCP)., Claude Cowork Runs a Real Python Sandbox. I Put 37 Million Rows of Mutual Fund NAV Data Through It., QRep - Powered by QuantStats. Live Now."
},
{
"slug": "quick-deploy-advanced-analysis-multi-agent-with-flowise",
"title": "Quick Deploy Advanced Analysis Multi-Agent with Flowise",
"tags": [
"database-ai"
],
"content": "Quick Deploy Advanced Analysis Multi-Agent with Flowise\n\nTIGZIG AI Co-Analyst app integrates reasoning-based analysis-with Deepseek, Gemini, Sonnet 3.7-into a multi-agent system, with Flowise for AI agent backend (Try it 👉 www.tigzig.com).\n\nFull-featured deploy can get complex with multiple repos and configurations...but here's a faster way to get started...\n\n4 step deploy (see video)\n\nFlowise provides a ready-to-go UI ...use it often in my client deploys.\nImport agent schemas\nUpdate Credentials.\nDeploy a SQL connector FastAPI server\nAdjust FastAPI security settings\n\nTop Tips\nNeed a free database? neon.tech (500MB/Postgres) spins up instantly. Options: Aiven (5GB/Postgres/MySQL), Supabase (500MB/Postgres)\nFile transfers to DB? DBeaver is a great option\nAdd a new reasoning model (e.g. Qwen-QwQ)? Replicate Deepseek agent & replace model name with OpenRouter's tag for Qwen\nCustomize- e.g. need an analysis plan without execution? Delete executor agent\nNeed new scenario? Modify router agent to add conditions and new routing.\nWhat about errors, validations, and latencies? -check out detailed post and video (link in last section)\nGet started with free credits: from ElevenLabs - 50,000 pages with Firecrawl, $50 at Render, $50 with Exa, and more: https://www.aiengineerpack.com/\n\nSource Codes& Flowise Agent Setups\n\nHands-On Perspective\n\nReal-world AI isn't just plug-and-play. In my projects (for SMBs), the hard part isn't spinning up an agent-it's data cleaning, validations, building usable data marts, automating processes, API flows, error handling, customizations as well as lots of agent iterations. I don't expect guides like these to solve everything, but do hope they provide a good starting point and insights\n\nBuild AI Co-Analyst Apps for Analytics\n\nExplore 15+ open-source AI analytics apps at tigzig.com with natural language to SQL, python charts, stats, database connectivity and more. Free access, source codes and build guides\nConnect ChatGPT to Supabase in 10 minutes: https://lnkd.in/g2A9h8f2\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, QDesk - Quant Report Desk\n\nExplore: Database AI Apps, MCP, Browse Apps\n\nMore posts: Flowise is my goto platform for GenAI and LLM apps, Multi-Agents (Sequential) with Reasoning – Connect to any database - o3-mini / Deepseek-R1 / Flash-2.0. Built with Flowise., AI Driven Advanced Analytics. Reasoning based Sequential Agents. Connect to any database — o3-mini/deepseek-r1 / gemini-flash-2.0., AI Co-Analyst — Live Multi-Agent App. Cost, quality, reliability — what works? what doesn’t?"
},
{
"slug": "qwen3-max-now-live-on-dats-4-sql-agent-suite-for-advanced-analysis-better-than-deepseek-r1-closer-t",
"title": "Qwen3 Max now live on DATS-4 SQL Agent Suite for Advanced Analysis Better than DeepSeek R1, closer to Claude Sonnet 4 - at a lower cost.",
"tags": [
"database-ai",
"text-to-sql"
],
"content": "Qwen3 Max now live on DATS-4 SQL Agent Suite for Advanced Analysis. Better than DeepSeek R1, closer to Claude Sonnet 4 - at a lower cost.\n\nQuality (my test scores)\nClaude Sonnet 4: 100\nGPT-5: 95\nQwen3 Max: 90\nDeepSeek R1: 85\nGemini 2.5 Flash: 75\n\nCost per 100 advanced analysis questions (USD)\nGemini 2.5 Flash: $14.25\nQwen3 Max: $14.5\nDeepSeek R1: $15.75\nClaude Sonnet 4: $19\nGPT-5: $27.5\n\nCost drivers\n\nEach advanced question = 7–10 SQL queries / tool calls. All executions run on GPT-4.1 (approx. $12.5 per 100), which is the major cost chunk. For simple one-shot queries, costs are negligible ($0.25 - $3.5 per 100)\n\nWorking rules (current)\nStick to OpenAI for execution\n4o-mini / 4.1-mini for simple & mid queries\n4.1 for complex executions\nSonnet 4 for top-end reasoning\nQwen3 Max, DeepSeek R1, GPT-4.1 for best value\nGemini 2.5 Flash if speed matters\nAvoid GPT-5 & Gemini Pro - very high cost + token bloat.\n\nTask example tested\nCredit card data mart build (1M customer, 10M transaction tables). Multi-step derivations, innovative derived variables creation, summarizations, merges. Table on AWS RDS MySQL.\nWeighted Score-Based Ranking of Indian credit card issuers - RBI Monthly File. Table on Supabase\n\nDATS-4\n\nFull-stack Database SQL Multi-Agent App: text-to-SQL, Python charts, statistical analysis, instant Postgres DB creation, PDF outputs, 9 reasoning models.\n\nLive app, source code, field guide\n\nwww.tigzig.com → Database AI & SQL Apps → DATS-4\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, BRIQ, Database Query Tool\n\nExplore: Database AI Apps hub, API/MCP hub, Guides hub\n\nMore posts: Sonnet 4.5. Released yesterday. Now live on DATS-4 SQL Agent Suite. Solid upgrade, but more 4.2 than 4.5., LLM Costing for Database AI Apps. Live Experience. Live App. Open Source, Database AI & SQL - Now choose you LLM: GPT-5, Deepseek, Qwen 3 Thinking. Live. Open Source."
},
{
"slug": "real-time-voice-ai-from-cricket-to-credit-cards-live-app-open-source",
"title": "Real-time voice AI - from cricket to credit cards. Live app. Open source.",
"tags": [
"voice-ai"
],
"content": "Real-time voice AI - from cricket to credit cards. Live app. Open source.\n\nReal-time Voice AI\n\nReal-Time Voice AI - From Cricket to Credit Cards 👉 Try it: tigzig.com/rt-cricket Mobile Friendly. Open Source. Eleven Labs\n\nBuilt with Eleven Labs Conversational AI rapid deploy voice widget.\n\n20+ years of ODI Cricket data. 1.5M rows in PostgreSQL.\n\nTesting realtime conversational AI - querying and DB-backed conversations:\n\nREX-C is a database-connected AI assistant built with ElevenLabs for real-time voice-to-SQL and more. While the demo uses cricket data, it can supports domains like finance, retail and beyond.\n\nRealtime Options - tested with live working prototypes\nOpenAI Realtime API (REX-R): ~10c/min (GPT-4 Turbo Mini) + WebRTC. Gold standard. Needs full coding; starter kits available. REX-R V2 coming soon.\nElevenLabs (REX-C): ~20c/min. Easy setup, rapid deploy, full SDK/API for customization.\n\nOther Options:\nHume AI: ~7c/min. Expression analysis, Next.js templates.\nVAPI: ~5c/min + actuals. SDK/API.\nOthers: Google Gemini Flash 2.0, Bland AI, Synthflow AI (~10c/min).\n\nPricing varies by usage. Most support inbound/outbound call integrations.\n\nSource Code/ Tech\n\nLightweight UI (~80kb Vanilla JavaScript, single code file). FastAPI for DB, ElevenLabs for voice, Flowise AI for LLM, Make.com for workflows.\n\nCaveat:\n\nWorking prototype. Single agent. Data is unvalidated, sourced from Cricsheet.org's experimental CSV/ZIP\n\nCricket data from Cricsheet, published under the Open Data Commons Attribution License 1.0 (ODC-BY).\n\n\nRelated\n\nTools: Voice DB - ElevenLabs, Voice DB - OpenAI Realtime, Cricsheet File Processor\n\nExplore: MCP Hub, Database AI Apps, Build Guides\n\nMore posts: Realtime voice AI - OpenAI WebRTC Implementation. Live app. Open source., Cricket (ODI/T20) & Tour de France stats from a Custom GPT connected to 3 live databases., Building AI apps with natural language and voice: top 9 tips"
},
{
"slug": "realtime-voice-ai-openai-webrtc-implementation-live-app-open-source",
"title": "Realtime voice AI - OpenAI WebRTC Implementation. Live app. Open source.",
"tags": [
"voice-ai"
],
"content": "Realtime voice AI - OpenAI WebRTC Implementation. Live app. Open source.\n\nREX-RT Voice AI\n\nThe real-time AI space is moving fast. Exploring options for real-time, conversation-driven analytics and DB-backed interactions.\n\nOpenAI's revamped Realtime API, Google's Flash 2.0 Experimental, and ElevenLabs' Conversational AI are all raising the bar while driving down costs.\n\nBuilt REX-RT, a mobile-friendly app powered by OpenAI's Realtime API. It connects to a PostgreSQL database with 1.5M cricket records-live, searchable, and fast. Using WebRTC, gpt-40-mini, and function calling. Costs ~$0.05–$0.07 per minute. Fully customizable.\n\nWhile the demo uses cricket data, it can support domains like finance, retail and beyond.\n\nPerformance: Solid, but not plug-and-play. Needs deep dives into event flows, WebRTC, and APIs. It's a code-and-build journey.\n\nLast week, I shared REX-C (cricket-odi.tigzig.com), built with Eleven Labs' Realtime Voice Widget. Plug-and-play simplicity, excellent performance (~$0.20/min), but limited UI customization-unless you use their SDK.\n\nThe possibilities: Huge. Conversation-driven analytics, voice integrations, realtime DB backed conversations, scalable apps for enterprises.\n\nOther tools to explore: Gemini 2.0 Flash Experimental, Eleven Labs, Hume AI and packaged ones like Vapi, Bland AI, Synthflow and others.\n\nTech Details-Source Code\n\nREX-RT is built with Cursor AI. Vanilla JS/HTML with CSS for mobile responsiveness, FastAPI for DB connectivity, and Flowise AI for LLM agents. Basic security is via domain whitelisting.\n\nMost of my apps lean on React, but loving Vanilla JS for its raw power and speed to deploy-though it trades some polish for simplicity.\n\nSource code, deployment guides and learning resources: tigzig.com \n\nCaveat: This is a working prototype with a single agent (gpt-40-mini). Data is sourced from cricsheet.org's experimental CSV section and isn't independently validated.\n\nCricket data from Cricsheet, published under the Open Data Commons Attribution License 1.0 (ODC-BY).\n\n\nRelated\n\nTools: Voice DB - OpenAI Realtime, Voice DB - ElevenLabs, Database AI Apps\n\nExplore: MCP / API hub, Browse all apps, Guides hub\n\nMore posts: Real-time voice AI - from cricket to credit cards. Live app. Open source., Meet REX-1: Your Realtime AI Analytics Agent System (Web Version), VOICE MODE - Querying & Analyzing Data with Custom GPT AWS - Azure Data Warehouse"
},
{
"slug": "related-party-transactions-vigil",
"title": "Related Party Transactions - Now Live on VIGIL",
"tags": [
"vigil"
],
"content": "Related Party Transactions - Now Live on VIGIL\n\nsrc: /images/blog/relatedParty01.png\nalt: Related Party Transactions on VIGIL\n\n250,000 transactions across 728 companies and 1,440 filings.\n\nExtracted from half-yearly XBRL filings (Sep 2025 + Mar 2025) filed under SEBI LODR Reg 23.\n\nCovers Nifty Total Market universe - 728 of 750 companies, 22 pending.\n\nEvery transaction pulled from the RPT section of standalone financials XBRL. Full extraction validation: each filing checked 3 ways - raw XBRL context count (independent of parser) vs parser output vs database, boolean flag cross-check, and field completeness audit. Zero failures across all 1,440 filings.\n\nThe relationship field in filings is free text - every company describes it differently. Built a text parsing logic that groups them into 12 standardized categories - Group Companies, KMP, Promoter Group, Relatives, Common Control etc.\n\nFilter by transaction type (loans, investments, remuneration, ICDs, sales, purchases), relationship group, company, Nifty indices. Copy and CSV export.\n\nSits alongside ratings, insider trading, pledges, encumbrance, takeover and surveillance on the company page.\n\nFree to use. No Ads. No login. Detailed Docs. CSV Downloads.\n\nvigil.tigzig.com\n\n\nRelated\n\nTools: VIGIL - India Red Flag Events, QRep - Security Reports, Quants Agent\n\nExplore: Browse all apps, Open data APIs and MCP, Analysis library\n\nMore posts: Evaluating a stock? Pull its VIGIL Report. Free., Is your company's promoter pledging shares to raise money? Are lenders releasing the pledge - or invoking and taking control?, New on VIGIL: SAST Takeover Disclosures (India), New feature on VIGIL: Rating Red Flags (India), VIGIL Data API and Download - 315K+ Records, 7 Tables, Updated Daily"
},
{
"slug": "releasing-brsr-analytics-india-esg-workforce-wages",
"title": "Releasing BRSR Analytics Tool. For Tracking Indian Corporate ESG, Workforce, Wages, Complaints.",
"tags": [
"vigil"
],
"content": "Releasing BRSR Analytics Tool. For Tracking Indian Corporate ESG, Workforce, Wages, Complaints.\n\nPublished: 2026-04-30\n\nBRSR Analytics Tool released on VIGIL. SEBI BRSR XBRL data for FY 2024-25 covering 712 Nifty Total Market companies. ~1.25 crore workforce. Over Rs.8 lakh crore wages disclosed. 8 indices, 22 sectors. Aggregate views, company drilldowns, benchmarking, XBRL viewer, data explorer, CSV download, separate DQ tab for data quality issues.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/releasing-brsr-analytics-india-esg-workforce-wages\nMarkdown of the HTML page: send GET https://tigzig.com/post/releasing-brsr-analytics-india-esg-workforce-wages with header Accept: text/markdown\n\nTags\nvigil\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nINDIA INC. · FY 2024–25\n\nWorkforce,\n\nWages & Complaints.\n\nA new analytics tool for India's corporate ESG disclosures - 712 listed companies, ~1.25 crore workers, ₹8 lakh crore in wages.\n\nWorkforce\n\nWages\n\nWage gap\n\nHuman rights\n\nSexual harassment complaints\n\nAmar Harolikar · Decision Sciences & Applied AI\n\nReleased 02 MAY 2026 · Sourced from SEBI BRSR XBRL filings\n\nTIGZIG\n\nLIVE NOW · VIGIL → BRSR\n\nFrom 712 XBRL fililngs to one workspace. Aggregate, compare, benchmark.\n\nvigil.tigzig.com → BRSR\n\nTIGZIG\n\nvigil.tigzig.com → BRSR\n\nCross-company analytics on BRSR XBRL filings\n\nAggregate views across 712 Nifty Total Market filers.\n\nCompare any set of companies side-by-side.\n\nXBRL viewer - inspect raw filings.\n\nData explorer - chart any metric.\n\nCSV download - export the dataset.\n\n'Report' button to flag issues.\n\nvigil.tigzig.com/brsr\n\nAmar Harolikar · Decision Sciences & Applied AI\n\nvigil.tigzig.com\n\nCompany view - one company vs. its sector or index benchmark.\n\nTIGZIG\n\nvigil.tigzig.com → BRSR\n\nIndia Inc. workforce in one frame\n\nFY 2024–25 BRSR filings, Nifty Total Market. Numbers on one chart you would otherwise reconstruct from 712 PDFs.\n\nFILERS\n\n712\n\nBRSR filers from the Nifty Total Market index, FY25 reporting cycle.\n\nTOTAL WORKFORCE\n\n~1.25Cr\n\nAggregate headcount disclosed across employees and workers, all categories.\n\nON CONTRACT\n\n~⅓\n\nRoughly one-third of the disclosed workforce is non-permanent - contract employees and contract workers.\n\nFEMALE SHARE\n\n<⅕\n\nLess than one-fifth of total disclosed headcount across India Inc.\n\nWAGES DISCLOSED\n\n₹8LCr+\n\nTotal wages disclosed in BRSR filings - over eight lakh crore rupees.\n\nINDICES\n\n8\n\nSlice across 8 indices and 22 sectors - and intersect them.\n\nSECTORS\n\n22\n\nSector cuts spanning IT, BFSI, Pharma, Auto, Metals, FMCG and more.\n\nREPORTING PRINCIPLES\n\nA + P5\n\nBRSR Section A plus Principle 5 (human rights). More principles to follow.\n\nAmar Harolikar · Decision Sciences & Applied AI\n\nvigil.tigzig.com\n\nTIGZIG\n\nvigil.tigzig.com → BRSR\n\nWhat you can do with it\n\nSix surfaces, one dataset. Every view links back to the source XBRL.\n\n01 / Aggregate\n\nCross-company distros\n\nSee the full distribution across 712 filers - wage gap buckets, complaint rates, female representation, headcount mix.\n\n02 / Compare\n\nSide-by-side\n\nPick any set of companies and put their workforce, wage and complaints disclosures next to each other.\n\n03 / Benchmark\n\nCompany vs. peer\n\nPlace a single company against its sector or index benchmark - including the rate-bucket it sits in.\n\n04 / Explorer\n\nDrill from chart to filing\n\nEvery aggregate clicks through to company-level numbers and raw line items.\n\n05 / XBRL viewer\n\nRead the source filing\n\nInspect the original XBRL in-browser. Full CSV download for any view.\n\n06 / Data Quality\n\nData Quality Analytics\n\nWrong units, copy-pasted values, mismatches surfaced as their own dashboard. Report button on every view.\n\nAmar Harolikar · Decision Sciences & Applied AI\n\nvigil.tigzig.com\n\nSECTION · FROM BUILDING IT\n\nThree things that surprised me.\n\n01\n\nCompanies with large female workforces reporting zero POSH complaints.\n\n02\n\nThe wage gap in leadership looks more severe than at the worker level.\n\n03\n\nData quality was worse than any regulatory filing I have seen - bad enough to need its own tab.\n\nTIGZIG\n\nvigil.tigzig.com → BRSR\n\nThe zero-complaint puzzle\n\nAcross 712 filers, 418 (59%) reported zero POSH complaints in FY25 - many of them with sizable female workforces.\n\nTHE QUESTION\n\nEither these companies are doing something the rest of India Inc. should learn from, or there is a reporting gap.\n\nAggregate filing rate sits at 12.38 POSH complaints per 10K female workforce across the Nifty 750. The zero-bucket dwarfs every other bucket - 418 of 712 filers.\n\nPOSH filing rate distribution · FY25\n\nCompanies bucketed by complaints filed per 10K female workforce\n\nRATE / 10K FEMALE\n\nCOMPANIES\n\nSHARE\n\n0\n\n418\n\n59%\n\n0–5\n\n27\n\n4%\n\n5–10\n\n41\n\n6%\n\n10–25\n\n92\n\n13%\n\n25+\n\n131\n\n18%\n\nNo female workforce\n\n1\n\n0%\n\nData quality flagged\n\n2\n\n0%\n\nSource: SEBI BRSR XBRL filings, FY25 · 712 Nifty Total Market filers\n\nAmar Harolikar · Decision Sciences & Applied AI\n\nvigil.tigzig.com\n\nTIGZIG\n\nvigil.tigzig.com → BRSR\n\nWage gap widens at the top\n\nMedian female-male wage gap across 712 BRSR filers, FY25. Steeper at KMP and Directors than at employees - the opposite of what most readers assume.\n\nPERMANENT EMPLOYEES\n\nThe baseline\n\n−7%\n\nmedian wage gap\n\nMost filers sit between −25% and ±5%. Real, but bounded.\n\nKEY MANAGERIAL PERSONNEL\n\n9× steeper than employees\n\n−64%\n\nmedian wage gap\n\n21% of filers below −50%. 60% report all-male KMP entirely.\n\nBOARD OF DIRECTORS\n\n7× steeper than employees\n\n−47%\n\nmedian wage gap\n\n17% of filers below −50%. Distribution far more dispersed than employees.\n\nAmar Harolikar · Decision Sciences & Applied AI\n\nvigil.tigzig.com\n\nTIGZIG\n\nvigil.tigzig.com → BRSR\n\nData quality, the elephant in the room\n\nDQ was bad enough that most of the early \"bugs\" in the app turned out to be filer errors. So the app got its own DQ tab with 17 distinct DQ categories & a DQ Explorer\n\n×\n\nWrong units. Numbers reported in lakhs, crores and units interchangeably across the same filing.\n\n×\n\nCopy-pasted values. Identical figures across rows that should not be identical - across years, categories, genders.\n\n×\n\nInternal totals not tallying. Disclosed sub-categories that don't add up to the disclosed total in the same filing.\n\n×\n\nPDF vs XBRL mismatches. Same disclosure but different numbers across the two formats from the same company.\n\n\"Worse than any other regulatory filing I have analyzed - and I have looked at a few.\"\nObservation while building the BRSR Analytics tool\n\nAmar Harolikar · Decision Sciences & Applied AI\n\nvigil.tigzig.com\n\nTIGZIG\n\nvigil.tigzig.com → BRSR\n\nA NOTE TO MY FRIENDS ON THE COMPANY SIDE\n\nBRSR is one tough reporting. I have been on that side too.\n\nCOMING SOON\n\nA free, browser-based DQ tool for BRSR filers. Runs locally in the browser - nothing leaves the machine.\n\nWHY\n\nHonestly self-interest. Cleaner filings make the analytics a lot easier to build.\n\nAmar Harolikar · Decision Sciences & Applied AI\n\nvigil.tigzig.com\n\nLIVE NOW · RELEASED 02 MAY 2026\n\nOpen the BRSR app.\n\nIndia VIGIL App is the home for BRSR Analytics. Aggregate, compare, benchmark, and read the raw XBRL - all from one place.\n\nvigil.tigzig.com\n\n→\n\nBRSR\n\nCOVERAGE\n\n712 BRSR filers Nifty Total Market · FY25\n\nSURFACES\n\nAggregate · Compare · Benchmark Explorer · XBRL · Data Quality\n\nFOUND A BUG?\n\nUse the Report button on any view.\n\nAmar Harolikar · Decision Sciences & Applied AI\n\nTIGZIG\n\nvigil.tigzig.com → BRSR\n\n\nRelated\n\nTools: VIGIL - India Red Flag Events, TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: BRSR Analytics deck (PDF), Open data APIs and MCP, Browse all apps\n\nMore posts: India Inc's BRSR Filings - All 656 in One Analytics Workspace. Wage Gap at Top Management: 68%. 1 in 3 Workers on Contract., Published Full Validation Reports for Every Dataset on VIGIL, VIGIL Data API and Download - 315K+ Records, 7 Tables, Updated Daily, VIGIL - India Red Flag Events Tracker v2 Release, India Was the Market's Darling. Not Any More. The Growth Story Has a Big Red Flag."
},
{
"slug": "releasing-mdrift-mutual-fund-composition-drift-analytics",
"title": "Releasing MDRIFT - Mutual Fund Composition & Drift Analytics Tool",
"tags": [
"mutual-funds"
],
"content": "Releasing MDRIFT - Mutual Fund Composition & Drift Analytics Tool\n\nMDRIFT\n\nPain point - MF holdings are monthly Excel files - different formats across AMCs, blank ISINs, corporate actions, CDs and CPs and T-Bills all over the place.\n\nComparing holdings across funds and time periods... huge pain...\n\nMDRIFT automates all of that. I download and process the files, standardize ISINs, handle corporate action remaps, group debt instruments... with validations carried out at each stage... and put it in a clean interface where you can drill down from scheme type to fund to individual security at ISIN level.\n\nCurrently covering 21 equity funds - top funds across Large Cap, Mid Cap, Small Cap, Flexi Cap and Focussed categories. Sept 2025, Dec 2025 and Jan 2026.\n\nQuick findings, Dec '25 vs Jan '26\nSmall Caps: rising interest in MCX\nParag Parikh Flexi Cap: added SBI, increased TCS\nMotilal Oswal Mid Cap: exits Polycab India, enters BEL, IDFC First\n\nSept '25 vs. Jan '26\nSmall & Mid Caps: rising interest in MCX, BSE Ltd, falling in Dixon Tech, Eternal (Zomato), Kaynes Tech\n\nTry it\n\nApp live at tigzig.com/mf-portfolio-dashboard\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, MF File Converter, MF Holdings Analyzer\n\nExplore: Mutual fund tools hub, Quant apps hub, Analysis deck\n\nMore posts: MF Composition Analytics With MDRIFT - Interesting Moves in Top Flexi Cap and Focused Funds, How I Identify and Map Every Holding — The ISIN Mapping Process, Analysis-as-App: Inside India’s Top Midcap Funds: Buys, Sells, Entries and Exits. Interactive Dashboard Release (Analysis-as-App), Now live - MFPRO v2 - Mutual Fund Analytics (India) - now with rolling returns, custom eval periods, multi-period & multi-instrument comparisons, Monthly MF portfolio files = hours wasted re-formatting. Here’s a tool that fixes it"
},
{
"slug": "releasing-module-02-practitioner-s-series-on-xlwings-lite-python-in-excel-data-cleaning-rule-b",
"title": "Releasing Module 02 — Practitioner’s Series on xlwings Lite. Python in Excel. Data Cleaning & Rule Based Transformation",
"tags": [
"xlwings-lite",
"python-in-excel"
],
"content": "Releasing Module 02 - Practitioner's Series on xlwings Lite. Python in Excel. Data Cleaning & Rule Based Transformation\n\nI've put up a new guide on my site: Module 02 of my practitioner's series on xlwings Lite, focused on data cleaning.\n\nMy go-to for any Python-in-Excel work is xlwings Lite from Felix Zumstein. The guide uses a real-world mobile number example to show how to handle multi step rule-based transformations with xlwings Lite.\n\nThis same methodology can be used for data cleaning of other variables. As well create complex derived variables- from analysis datasets to campaign files.\n\nBut it's really about the process: how to instruct AI, how to use AI Coder file to provide context, and most importantly, validation of output.\n\nThis brings me to a core lesson from my years in the trenches: Validations are sacrosanct. AI didn't do the work- you did. It's a tool that helps, that's it. If it's not validated, it does not go out. No matter what AI says. No matter the pressure. Period.\n\nGet the full module- including the workbook and AI Coder file at Tigzig xlwings Lite section\n\nxlwings-lite.tigzig.com\n\nxlwings Lite official site: lite.xlwings.org\n\n\nRelated\n\nTools: Practice Lab, DUCKIT - CSV to DuckDB, Data Importer\n\nExplore: AI in Excel hub, All Guides, Browse all apps\n\nMore posts: The xlwings Lite AI Coder Instruction File - December 2025 Release, Tool: A 1,450-line context file. Purpose: To ensure clean, efficient xlwings Lite code generation., xlwings Lite new WINGMAN function - some usage patterns: python sandbox, stats, cleaning, bucketing, judging"
},
{
"slug": "releasing-rex2-ai-decision-intelligence",
"title": "Releasing REX-2: AI Decision Intelligence",
"tags": [
"database-ai",
"text-to-sql"
],
"content": "Releasing REX-2: AI Decision Intelligence\n\nTry it Free: rex.tigzig.com (Mobile/Web). Open Source.\n\nFunctionalities\nNatural Language (NL)-to-SQL: Query, Append, Join, Merge, Summarize, Transform\nComplex Analysis: Cards Segment Profiling example\nNL-to-Python: Statistical Analysis & Python Charts\nInteractive Tables: Sort, Filter, Stats (Central Tendency, Dispersion)\nData Structure Reports & Quick AI Analysis\nPDF Generation: Formatted Reports\nConnect to ANY DW (MySQL/PostgreSQL)\nCreate Temporary DBs on the fly for rapid analysis\nData Sources: DB tables or Upload CSV/TXT (comma/pipe-delimited)\nAuthentication: Auth0\n\nTry it for Free\n\nrex.tigzig.com - No API key needed. CSV/TXT files available. Link in 'Help' section on site. Currently optimized for web/desktop only.\n\nSource Codes\n\n4 GitHub repos plus schemas. Links in the 'Build' section on the website + detailed video guide.\n\nVideo Guide\nUsing REX-2 Features: ~30 mins\nComplex Queries: ~30 min\nBuild Guide: ~90 min\nSee comments for timestamped TOC\n\nCaveats\nPrototype (working version)\nSingle LLM agent: stumbles occasionally. Full team on the way\n\nPlanned Features\nLLM / Agent Network: Maker/Checker/Work Segregation\nSQL-Python Log Streaming in UI, File downloads\nCustom Python Code Interpreter (PCI)\nNL-to-Python - large datasets (needs PCI)\nAdvanced Statistical Analysis (needs PCI)\nGeneric Excel to CSV Converters, Universal Web Scrapers\nReal-Time Voice (OpenAI Alt)\nSecurity / Frontend-Backend Segregation\nML Models (need PCI+)\nOn-the-fly Interactive Dashboards (Claude-like)\n\nTip: Setup Your Own Warehouse, Instantly\n\nNeon, Aiven, Supabase offer free tiers for production-ready databases. Set up in under a minute with just a few clicks.\n\nTip: Voice-to-SQL\n\nVoice-typed using Windows Voice (Win+H). Google Docs Voice Typing is also great. Longest prompt (~5 mins of voice instructions) converted into SQL (check out at YouTube timestamp 00:49:20).\n\nBuild Details\nBuilt with Cursor AI: My #1 AI-coding tool; recommend Volo's channel on YouTube. Links in 'Build' section\nAI Backend: FlowiseAI (API-connected). My top LLM App platform\nPython Charts/Stats: E2B on Flowise\nUI: React, TypeScript, Vite, Shadcn\nDB Connections: Custom built FastAPI servers\nDeployment: Vercel (main site), Hetzner (via Coolify for FastAPI), Render (Flowise)\nUser-Created DBs: Neon (blazing fast, great APIs, ideal for LLM apps)\n\nREX is an independent, open-source project I've developed as part of my desire to make Generative AI tools practical and accessible. It's also my way of sharing with the broader community, inspired by open-source platforms like Flowise that have benefited me immensely. Updates will roll out as I balance this with other projects.\n\n\nRelated\n\nTools: BRIQ, DATS-4 Database AI Suite, Database Query Tool\n\nExplore: Database AI Apps hub, APIs and MCP, Browse all apps\n\nMore posts: REX-2: Your AI Analyst on Call, Meet REX-1: Your Realtime AI Analytics Agent System (Web Version), Flowise is my goto platform for GenAI and LLM apps, Mistakes I Made Building Text-to-SQL Agents in Live Projects. My 2025 Learnings"
},
{
"slug": "releasing-us-credit-fdic-nyfed-interactive-tool",
"title": "Releasing US-CREDIT. For Analysts Tracking US Banking. Interactive Tool Covering FDIC SDI and NY Fed CCP.",
"tags": [
"portfolio-quants"
],
"content": "Releasing US-CREDIT. For Analysts Tracking US Banking. Interactive Tool Covering FDIC SDI and NY Fed CCP.\n\nPublished: 2026-04-20\n\nUS-CREDIT interactive analytics tool. 25 years, 100 quarters. $13.6T bank loans across 4400+ FDIC institutions. $18.8T consumer debt via NY Fed CCP. 18 FDIC segments, delinquency buckets, charge-offs, FICO bands, age groups. Validated against FDIC QBP - 100 quarters reconciled. Includes disbelief rally research deck.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/releasing-us-credit-fdic-nyfed-interactive-tool\nMarkdown of the HTML page: send GET https://tigzig.com/post/releasing-us-credit-fdic-nyfed-interactive-tool with header Accept: text/markdown\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nFDIC Q1 2026 - Bank NCO by Segment\n\nTIGZIG · US BANKS\n\nFDIC Q1 2026 · Net charge-offs by segment\n\nBanks holding the line ..\nbut the ‘lemons’ sit off the chart.\n\nNet charge-off rate (annualized) for US bank loans. Total at 0.58% sits right at pre-GFC.\nConsumer (15% of loans) has already crossed pre-GFC and is trending up.\nC&I has nudged just past. Real Estate and Other well below.\n\nTotal loans & leases\n\n$13.83T · 100%\n\nGFC peak\n\n2.89%\n\nPre-GFC\n\n0.56%\n\nQ1'26\n\n0.58%\n\nPrivate credit stress doesn't show up here. Mark-to-market hits P&L directly · does not flow through to NCO numbers.\n\nNY Fed consumer delinquencies have already broken pre-GFC. The weakest paper sits at credit unions and non-bank lenders.\n\nConsumer\n\n$2.05T · 15%\n\nGFC peak\n\n6.64%\n\nPre-GFC\n\n2.47%\n\nQ1'26\n\n2.87%\n\nCommercial & Industrial\n\n$2.57T · 19%\n\nGFC peak\n\n2.69%\n\nPre-GFC\n\n0.46%\n\nQ1'26\n\n0.52%\n\nSegment\nBalance · share\nGFC peak\nPre-GFC\nQ1'26\n\nReal Estate\n$6.16T · 45%\n2.50%\n0.22%\n0.05%\n\nAll Other*\n$3.05T · 22%\n1.41%\n0.24%\n0.14%\n\nQ1 2026 numbers now live\n\ntigzig.com → Tremor → US Credit\n\nBank loss & delinquency (FDIC) · consumer credit (NY Fed / Equifax) · available via UI with customizable charts, API, and MCP for AI agents.\n\nSource: FDIC quarterly bank aggregates (SDI / Call Reports), NSA, served via Tremor. · Pre-GFC reference: Q3 2007. · * All Other = agri loans, lease financing, all other loans, depository institutions, foreign gov't, state & local gov't.\n\nAmar Harolikar · Decision Sciences & Applied AI\n\ntigzig.com· tools for analytics, quants and macro signals\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, TREMOR Macro Indicators\n\nExplore: Private Credit Stress hub, Quant apps hub, Analysis decks\n\nMore posts: Banks Holding the Line.. But the 'Lemons' Sit Off the Chart. In Private Credit, Credit Unions, and Non-Bank Consumer Lenders., Q1 2026 US Lending Data Live on Tigzig. NCUA + FDIC + NY Fed in One API. Two of Three Not on FRED. AI-Agent Ready., Auto Losses Past the GFC Peak. Cards Almost There. Credit Unions Already at GFC Loss Rates. NY Fed + NCUA Data Now on Tigzig., The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top., The Crisis in Auto Loans. 2007 was Sub-Prime Mortgage. Is 2026 Auto?"
},
{
"slug": "rex-2-ai-driven-analytics-python-connect-to-any-database",
"title": "REX-2 : AI Driven Analytics",
"tags": [
"database-ai",
"text-to-sql"
],
"content": "Releasing REX-2: AI Decision Intelligence\n\nFunctionalities\nNatural Language (NL)-to-SQL: Query, Append, Join, Merge, Summarize, Transform\nComplex Analysis: Cards Segment Profiling example\nNL-to-Python\nStatistical Analysis and Python Charts\nInteractive Tables: Sort, Filter, Stats (e.g. Central Tendency, Dispersion)\nData Structure Reports and Quick AI Analysis\nPDF Generation: Formatted Reports\nConnect to ANY DW (MySQL / PostgreSQL)\nCreate Temporary DBs on the fly for rapid analysis\nData Sources: DB tables or upload CSV / TXT (comma or pipe-delimited)\nAuthentication: Auth0\n\nTry it for free\n\nAnalyzer \nNo API key needed. CSV / TXT files available. Link in the Help section on site. Currently optimized for web / desktop only.\n\nSource Codes\n\n4 GitHub repos plus schemas. Links in Docs on site.\n\nVideo Guide\n\nFull Video on YouTube\nUsing REX-2 features: approx. 30 mins\nComplex queries: approx. 30 mins\nBuild guide: approx. 90 mins\nSee comments for timestamped TOC\n\nCaveats\nPrototype (working version)\nSingle LLM agent: stumbles occasionally. Full team on the way.\n\nPlanned Features\nLLM / Agent Network: Maker, Checker, Work Segregation\nSQL-Python log streaming in UI, file downloads\nCustom Python Code Interpreter (PCI)\nNL-to-Python for large datasets (needs PCI)\nAdvanced statistical analysis (needs PCI)\nGeneric Excel to CSV converters, universal web scrapers\nReal-time voice (OpenAI alternative)\nSecurity and frontend-backend segregation\nML models (need PCI+)\nOn-the-fly interactive dashboards (Claude-like)\n\nTip: Setup your own warehouse instantly\n\nNeon, Aiven, Supabase offer free tiers for production-ready databases. Setup in under a minute with just a few clicks.\n\nTip: Voice-to-SQL\n\nVoice-typed using Windows Voice (Win + H). Google Docs voice typing also works well. Longest prompt (approx. 5 mins of voice instructions) converted into SQL. See YouTube timestamp 00:49:20.\n\nBuild Details\nBuilt with Cursor AI (my #1 AI-coding tool). Recommend Volo's YouTube channel. Links in Build section.\nAI Backend: FlowiseAI (YC S23), API-connected. My top LLM app platform.\nPython Charts and Stats: E2B on Flowise\nUI: React, TypeScript, Vite, Shadcn\nDB Connections: Custom-built FastAPI servers\nDeployment: Vercel (main site), Hetzner via Coolify (FastAPI), Render (Flowise)\nUser-created DBs: Neon (fast, strong APIs, ideal for LLM apps)\n\nTigzig is an independent, open-source project developed to make generative AI tools practical and accessible. It is also a way to give back to the community, inspired by platforms like Flowise that have been valuable to me. Updates will roll out as I balance this with other projects.\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, BRIQ, ChatGPT to Any Database\n\nExplore: Database AI Apps, MCP, Browse Apps\n\nMore posts: Database AI, built for day-to-day work. Five categories, ten micro apps. Live, open source, free., BRIQ App: DuckDB AI in Browser - 500MB Files, 4M+ Records, No Database Setup, Mistakes I Made Building Text-to-SQL Agents in Live Projects. My 2025 Learnings"
},
{
"slug": "rex-2-your-ai-analyst-on-call",
"title": "REX-2: Your AI Analyst on Call",
"tags": [
"database-ai",
"text-to-sql"
],
"content": "REX-2: Your AI Analyst on Call\n\nREX-2 Analyst\n\n𝐘𝐨𝐮𝐫 𝐀𝐈 𝐀𝐧𝐚𝐥𝐲𝐬𝐭 𝐨𝐧 𝐂𝐚𝐥𝐥 - 𝐑𝐄𝐗-𝟐\nIn <150 seconds: Database on the fly, Upload a 100MB File, Voice-to-SQL + Python Chart \n\nSharing a quick demo to show just how easy REX is to use. \n\nStep 1. Create a database on the fly in seconds.\nStep 2. Upload a TXT/CSV file (pipe or comma delimited).\nStep 3. Use voice chat to request analyses.\nStep 4. Get analysis tables and charts \n\n[The demo shows creating a new database on the fly. But you can as well connect to an existing warehouse and query your tables, or upload the text file there. Just needs the database credentials. Use the 'Connect' button for that]\n\n𝐕𝐨𝐢𝐜𝐞-𝐭𝐲𝐩𝐞𝐝 using built-in Windows Voice feature (Win+H). Mac users can use native voice dictation. If you prefer, Google Docs Voice Typing also works well. Of course, text-based queries work just as well if you’d rather type out your request.\n\n𝐑𝐄𝐗-𝟐 𝐨𝐟𝐟𝐞𝐫𝐬 𝐟𝐞𝐚𝐭𝐮𝐫𝐞𝐬 like Natural Language-to-SQL, interactive tables (with sorting, filtering, and key stats), statistical analysis, AI-generated reports, PDF generation. And seamless database connectivity-whether to your existing database or a temporary one created on the fly for rapid analysis.\n\n𝐁𝐮𝐢𝐥𝐝 𝐢𝐭 𝐨𝐮𝐭, customize and deploy with the source codes shared in the 'Build' section on the website + detailed video guide.\n\n𝐓𝐫𝐲 𝐢𝐭 for free at rex.tigzig.com. No API key needed. Feel free to use the CSV/TXT files available in the ‘Help’ section of REX. Currently optimized for web/desktop only.\n\n𝐕𝐢𝐝𝐞𝐨 𝐆𝐮𝐢𝐝𝐞\nUsing REX-2 Features: ~30 mins\nComplex Queries: ~30 mins\nBuild Guide: ~90 mins\n\n𝐃𝐞𝐭𝐚𝐢𝐥𝐞𝐝 𝐋𝐢𝐧𝐤𝐞𝐝𝐈𝐧 𝐏𝐨𝐬𝐭 (with features current and planned, caveats, build stack, tips and timestamped video table of contents)\nhttps://www.tigzig.com/post/releasing-rex2-ai-decision-intelligence\n\n𝑁𝑜𝑡𝑒: 𝑆𝑝𝑒𝑒𝑑𝑠 𝑚𝑎𝑦 𝑣𝑎𝑟𝑦 𝑏𝑎𝑠𝑒𝑑 𝑜𝑛 𝑓𝑖𝑙𝑒 𝑎𝑛𝑑 𝑞𝑢𝑒𝑟𝑦 𝑐𝑜𝑚𝑝𝑙𝑒𝑥𝑖𝑡𝑦, 𝑛𝑒𝑡𝑤𝑜𝑟𝑘 𝑏𝑎𝑛𝑑𝑤𝑖𝑑𝑡ℎ, 𝑎𝑛𝑑 𝑠𝑒𝑟𝑣𝑒𝑟 𝑝𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒.\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, Voice DB - OpenAI Realtime, BRIQ\n\nExplore: Database AI Apps, Browse All Apps, MCP Hub\n\nMore posts: Releasing REX-2: AI Decision Intelligence, REX-2 : AI Driven Analytics, Meet REX-1: Your Realtime AI Analytics Agent System (Web Version)"
},
{
"slug": "rex1-your-realtime-ai-analytics-agent-system-web-version",
"title": "Meet REX-1: Your Realtime AI Analytics Agent System (Web Version)",
"tags": [
"database-ai",
"text-to-sql"
],
"content": "Meet REX-1: Your Realtime AI Analytics Agent System (Web Version)\n\nBuilt on OpenAI's Realtime API. Connection to Data Warehouse, Text-to-SQL, Statistical Analysis, Web Scrape, Charts, Python Processing and Automation.\n\nText and voice-input options (non-realtime) also live.\n\nOpenAI recently released it advanced sound mode via its Realtime API. Wanted to test it out. REX-1 was the result\n\n▸ Web version. Best on full screen.\n\n▸ Realtime features need an API key. API Costs: ~$1/min.\n\n▸ Non-realtime with voice input: Free to use, but some features require DB credentials. Ignore the OpenAI key popup or enter random digits to close it. Click on the Task Agent or Yahoo Finance Agent to get started. Ask the agents what they can do, and you're good to go.\n\nREX-1 Capabilities\nConnect to Data Warehouses (AWS, Azure, MySQL)\nText-to-SQL, Data Analysis & Transformation\nStatistical Analysis & Python Charts\nWebsites Scraping\nStock Market Technical Charts & Data\nReporting Automation (Trackers, PDFs, Emails)\n\nYouTube Video\n\n'How to Build & Deploy Realtime API Analytics Assistant Agent System'. (90 mins). https://www.youtube.com/watch?v=99aD2tv8G-0\n\nSee time stamped table of contents below\n\nLearnings and insights\nPerformance is mind blowing and pricing blows your purse.\n$1 per minute or realtime , while text (with voice input ) less than $1 over hours of use.\nPrices may drop by a tenth in 6-9 months, giving early developers a first mover advantage\nWebSockets and other approaches offer near realtime alternatives\nIf use case allows, Custom GPTs very cost-effective; agents can be connected easily\nDeploy with various agent setups (sequential, specialized, master, backend etc). Architecture depends on the use case.\n\nBasic Build\n\nFrontend: TypeScript (OpenAI quick start repo)\n\nLLM Agents: Flowise AI. Agent API access, nested agents, code interpreter, FastAPI integration, rapid deployment, voice input. Open source with self-hosting option. FlowiseAI (YC S23) makes it incredibly easy to build and deploy complex apps quickly.\n\nWorkflow automation: Make.com.\n\nFastAPI Custom Servers: for DW connectivity and custom python processing\n\nSource Code JSONs, Schemas at: Tigzig\n\nAdditional\nFive-part detailed hands-on implementation guide to build AI Analytics Voice Based Assistant\n\nYouTube Time Stamps / Table of Contents\n00:04 - Introduction and Demo of tool capabilities\n02:01 - Agenda for video\n02:33 - Detailed Demo\n17:34 - Performance and Pricing Details\n24:37 - Architecture Overview\n\nFrontend, APIs, backend agent setup, tool by tool overview of all tools\n38:54 - Build Step 1: Setting up user interface\n44:45 - Build Step 2: Setup backend LLM Agents and Tools in Flowise\n48:18 - Database Tool\n51:57 - Reporting Automation Tool\n55:59 - Website Scraper Tool\n1:00:20 - Document Update Tool\n1:02:28 - Code Interpreter Tool\n1:05:28 - Yahoo Finance Agent\n1:11:04 - Google Custom Search Tool\n1:13:40 - Build Step 3: Connecting Backend to Frontend\n1:25:12 - Conclusion\n\nCaveats\n\nPrototype app. Working prototype.\n\n\nRelated\n\nTools: Voice DB - OpenAI Realtime, DATS-4 Database AI Suite, QDesk - Quant Report Desk\n\nExplore: MCP Hub, Database AI Apps, Quant Apps\n\nMore posts: Realtime voice AI - OpenAI WebRTC Implementation. Live app. Open source., REX-2: Your AI Analyst on Call, VOICE MODE - Querying & Analyzing Data with Custom GPT AWS - Azure Data Warehouse, Flowise is my goto platform for GenAI and LLM apps"
},
{
"slug": "rolling-returns-why-cagr-alone-can-mislead-you",
"title": "Rolling Returns: Why CAGR Alone Can Mislead You (And What To Use Instead)",
"tags": [
"mutual-funds",
"portfolio-analytics",
"duckdb"
],
"content": "Rolling Returns: Why CAGR Alone Can Mislead You (And What To Use Instead)\n\nThere is a common misunderstanding about CAGR. People hear \"Compound Annual Growth Rate\" and assume some kind of averaging is happening. It isn't. CAGR is purely point-to-point. Start NAV on date A, end NAV on date B, apply the compounding formula. That's it. One start date. One end date. One number.\n\nThe problem? That one number is fragile. Shift your start date by a week - especially around a market event - and your CAGR can look meaningfully different. A fund that shows 14% CAGR from January 3rd might show 11% if you start from January 10th. Same fund, same end date. Different number because the market had a bad week.\n\nFor shorter evaluation periods - 1 year, 2 years - this gets worse. Your entire assessment of a fund can swing based on whether the start date happened to land on a good day or a bad day. And when you're comparing two funds, both numbers are hostage to that same date sensitivity. You might pick the wrong fund simply because one of them got lucky with the calendar.\n\nThis is not a theoretical problem. It happens all the time.\n\nWhat Rolling Returns Fix\n\nRolling returns shift the question from \"what happened between these two specific dates\" to \"what does a typical holding period actually look like across the fund's entire history.\"\n\nHere is how it works. Take every single trading day in a fund's history as an end date. For each one, look back exactly 1 year and compute the CAGR for that window. Do this for every possible end date. Now instead of one number, you have thousands of individual 1-year CAGRs.\n\nA fund with 7 years of history will give you roughly 1,500 to 1,700 individual 1-year CAGR observations. Average those. Find the median. Find the worst. Find the best. Count how many were negative.\n\nThat average is a far more reliable picture of what a 1-year holding typically looks like. No single date can distort it. No lucky or unlucky start point can skew the result.\n\nTwo Periods That People Confuse\n\nRolling returns involve two distinct time spans. Getting them mixed up is easy - and common.\n\nRolling Window - this is the holding period being simulated. If you select \"1 Year\", each individual observation is a 1-year CAGR. Select \"3 Years\" and each observation is a 3-year CAGR. This answers: \"what would I have earned if I held for this long?\"\n\nEvaluation Period - this is the overall historical span across which those windows are computed. For example, January 2013 to March 2026. A longer evaluation period means more observations, and more observations means more reliable statistics.\n\nThe statistics you see - average, median, % negative - are summaries across all the rolling observations within the evaluation period.\n\nImportant: the evaluation period must be meaningfully longer than the rolling window. A 5-year rolling window needs at least 7-8 years of data to produce a useful distribution. A 3-year window needs at least 4-5 years. If a fund launched in 2022 and you select a 5-year window, there's barely enough history for a single valid observation. The number of observations shown in the results tells you how much data went into the calculation.\n\nHow the Computation Works\n\nA few technical details for those who want to go deeper.\n\nCalendar days, not trading days. A \"1 Year\" window looks back exactly 365 calendar days from each end date - not 250 trading days. Since markets are closed on weekends and holidays, the target date 365 days ago may not have a NAV. The system uses something called an ASOF JOIN - it finds the most recent available NAV on or before the target date. So if the target falls on a Saturday, it picks up the Friday NAV.\n\nThe CAGR is then computed over the actual calendar days between the two matched dates - not an assumed 365.\n\nThe formulas:\n\nFor 1-year and longer windows:\nCAGR = (End NAV / Start NAV) ^ (365.25 / Actual Days) - 1\n\nFor shorter windows (3 months, 6 months):\nAbsolute Return = (End NAV / Start NAV - 1) x 100\n\nWhy the split? Annualizing short periods can be misleading. A 5% return in 3 months annualized to ~21% sounds great but that's not what actually happened. So for sub-1-year windows, I show the actual absolute return.\n\nMinimum gap thresholds. Each window has a minimum gap requirement. The actual days between start and end NAV must exceed this minimum to be counted. This filters out edge cases where extended market closures cause the matched dates to land too close together.\n\n| Window | Look-back | Min Gap | Return Type |\n|--------|-----------|---------|-------------|\n| 3 Months | 91 days | 46 days | Absolute % |\n| 6 Months | 183 days | 92 days | Absolute % |\n| 1 Year | 365 days | 330 days | CAGR |\n| 3 Years | 1,096 days | 986 days | CAGR |\n| 5 Years | 1,826 days | 1,643 days | CAGR |\n\nThe ASOF JOIN. The whole computation runs in DuckDB SQL on the backend. For each trading day (end date), the ASOF JOIN finds the most recent NAV on or before the target start date in a single pass. This is what makes it efficient - computing rolling returns for all 95+ funds across all 5 windows takes under 10 seconds.\n\nHere is the simplified logic:\n\n``sql\nSELECT\n t.fund,\n t.date AS end_date,\n p.date AS start_date,\n POWER(t.nav / p.nav, 365.25 / daysbetween) - 1 AS rollingcagr\nFROM nav_daily t\nASOF JOIN nav_daily p\n ON t.fund = p.fund\n AND p.date <= t.date - INTERVAL 365 DAY\nWHERE days_between >= 330\n``\n\nIf you're not familiar with ASOF JOIN - it's a DuckDB feature that matches each row to the nearest row meeting the condition, without scanning everything. Regular JOINs need exact matches. ASOF finds the closest match. Very useful for time-series data where dates don't align perfectly.\n\nReading the Results - What Each Column Means\n\nIf you run a fund-vs-fund or fund-vs-index comparison in MFPRO, you get a rolling returns table with these columns. Here is what each one tells you:\n\nAverage CAGR - The mean of all rolling returns. Think of it as your \"expected\" return if you had invested on a random day within the evaluation period. More stable than point-to-point CAGR because thousands of observations go into it.\n\nMedian CAGR - The middle value. Less affected by extreme observations than the average. When the median is significantly lower than the average, the fund has some very high outlier returns pulling the average up - which means typical experience may be lower than the average suggests.\n\nMin CAGR & Date - The worst rolling return and when that window ended. Usually corresponds to investing just before a major crash. Tells you the worst case scenario for that holding period.\n\nMax CAGR & Date - The best rolling return and when it ended. Usually corresponds to investing at a market bottom and catching the recovery.\n\n% Negative - What percentage of all rolling windows produced a negative return. This is probably the most practical column. A fund showing 10% negative on 1-year rolling means roughly 1-in-10 chance of losing money if you held for a year, regardless of when you invested. A fund with near-zero % negative on 3-year rolling? Very consistent - you almost never lost money over any 3-year period.\n\nObservations - How many rolling windows were computed. More observations = more reliable statistics. If this number is low (say under 500 for 1-year rolling), treat the stats with caution.\n\nA practical way to use this: A fund with high average CAGR but also high % negative is high-return but inconsistent. A fund with slightly lower average but near-zero % negative is more reliable. Which one you prefer depends on your holding discipline and risk tolerance. Rolling returns make that trade-off visible in a way that a single CAGR number never can.\n\nHow to Validate This Yourself in Excel\n\nOne thing I'm particular about - every number on MFPRO should be independently verifiable. I don't want anyone taking my calculations on faith.\n\nGo to the Returns section in MFPRO, then the Validation tab. There you will find:\nRaw NAV and index data downloadable as CSV\nA pre-built Excel file with CAGR and rolling return formulas already set up\nStep-by-step instructions on how to replicate every number\n\nI ran this validation on BSE 100 Index data. 7,485 individual rolling return values across 3 windows and 3 evaluation periods. Every single value came within 0.50 basis points of the manually computed Excel result. The tiny differences (sub-basis-point) come from date matching - the ASOF JOIN picks the nearest trading day, which can differ by a day from a strict calendar lookup - and floating-point precision.\n\nThe validation detail:\n\n| Window | Data Points | Within 0.01 bps | Within 0.50 bps | Max Variance |\n|--------|-------------|-----------------|-----------------|--------------|\n| 1Y Rolling | 2,983 | 99.70% | 100.00% | 0.44 bps |\n| 3Y Rolling | 2,497 | 99.24% | 100.00% | 0.42 bps |\n| 5Y Rolling | 2,005 | 98.80% | 100.00% | 0.46 bps |\n| Total | 7,485 | 99.28% | 100.00% | 0.46 bps |\n\nDownload the files, plug in your own formulas, and check. If you find a discrepancy beyond 0.50 bps, I want to know about it.\n\nSee It Live\n\nAll of this is live on MFPRO. 95 funds across 6 categories, 9 indices. Compare rolling returns across funds and benchmarks, custom evaluation periods, data downloads for offline analysis.\n\nFull methodology docs with detailed examples are on the app.\n\nmfpro.tigzig.com\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, QRep - Security Reports, Quants Agent\n\nExplore: Mutual fund tools, Quant apps hub, Analysis library\n\nMore posts: Now live - MFPRO v2 - Mutual Fund Analytics (India) - now with rolling returns, custom eval periods, multi-period & multi-instrument comparisons, That 9X return from Nifty Midcap is irrelevant if you couldn't survive the 73% of time it was in drawdown, Gold up 2.3X past 2 years. But what about the 10 years drawdown in between., U.S. Markets (S&P 500) vs India (Nifty 50) - is the returns profile reversing?"
},
{
"slug": "run-a-full-ai-database-app-as-a-single-html-file-no-server-no-remote-db",
"title": "Run a Full AI Database App as a Single HTML File. No Server. No Remote DB.",
"tags": [
"database-ai",
"duckdb"
],
"content": "Run a Full AI Database App as a Single HTML File. No Server. No Remote DB.\n\nFew days back I shared a custom implementation powered by SQL Rooms (sqlrooms.org) to chat, query multi-GB files in conversational language. With DuckDB database running right in browser. Live here\n\nNow, you can download the full thing - one HTML file that works right off your laptop. The entire React application is now compiled into a single, portable 3.5MB HTML file.\n\nDouble-click it, and you're running a full AI app. No installs. No backend. No server deploys. Download it, email it, or share it.\n\nImported a 1.6GB / 11M-row files - chatted with the data to run analysis and visualizations (see video)\n\nTry it\n\nDownload live cricket data files (100MB - 1.6GB unzipped) from my Google Drive. Get a free Google Gemini API key with your Gmail account from aistudio.google.com, and have at it.\n\nUse cases\nFully portable apps. Run directly from your local drive.\nInternal tools or lightweight AI utilities\nShare by email or keep on shared drive\nReplace static decks with live, interactive apps\nWorks for AI, automation, or dashboards alike\n\nTechnical Note\nThe build uses Vite with the vite-plugin-singlefile plugin, which bundles all JS/CSS into the HTML. Code and methodology is in the repo.\nI have currently enabled only Google Gemini. CORS issue with OpenAI and Anthropic API calls but can be enabled with a simple proxy server.\n\nTo build\n\nUse Google Build (Free) or Lovable to setup your app and a working version. Then build locally with Claude Code, Cursor, or Gemini CLI and export via Vite singlefile plugin.\n\nSource Code\n\nDocs: On app → \"Docs\" tab\n\nCredits: Original SQL Rooms\n\nsqlrooms.org from Foursquare. Architected by Ilya Boyandin\n\nCricket data from Cricsheet, published under the Open Data Commons Attribution License 1.0 (ODC-BY).\n\n\nRelated\n\nTools: BRIQ, Cricsheet File Processor, DUCKIT - CSV to DuckDB\n\nExplore: Database AI Apps hub, Database Query Tool (MCP), Browse all apps\n\nMore posts: Chat, Query, and Transform Multi-GB Files - In Natural Language, Right in Your Browser with DuckDB., Segment 1M customers from 10M transactions (640MB CSV) with natural language queries / Text-to-SQL - entirely in your browser. No server. No remote database. No IT approvals., Bundle your AI app or React dashboard into a single file."
},
{
"slug": "run-advanced-analytics-locally-in-your-browser-no-server-no-remote-database-no-it-approvals",
"title": "Segment 1M customers from 10M transactions (640MB CSV) with natural language queries / Text-to-SQL - entirely in your browser. No server. No remote database. No IT approvals.",
"tags": [
"duckdb",
"text-to-sql"
],
"content": "Segment 1M customers from 10M transactions (640MB CSV) with natural language queries / Text-to-SQL - entirely in your browser. No server. No remote database. No IT approvals.\n\nProcess multi-GB files, build datamarts, and run AI-powered analysis entirely in your browser with DuckDB + text-to-SQL AI. When server deployment isn't an option - or will take 6 months and $1M in approvals - this runs off a single HTML file on your laptop.\n\nDABX-1 (Database AI, Browser, eXtended) - built on SQL Rooms (sqlrooms.org) by Ilya Boyandin, customized for analytics workflows that need to move fast without infrastructure headaches.\n\nLive at www.tigzig.com/sql-rooms. Or download the full app as a single 3.5MB HTML file and run it locally.\n\nTwo Core Value Propositions\nLocal Browser Execution = Data Privacy\n\nYour data never leaves your machine. Files import directly into DuckDB running in the browser. No uploads to remote servers. No cloud storage. API calls go to the LLM for query generation, but your actual data stays local.\nSingle-File Deployment = Zero Infrastructure\n\nWhere traditional deployment requires server provisioning, security reviews, and IT approvals - this bypasses all of it. Download one HTML file. Double-click. You're running a full AI-powered analytics app. Share it like you share Excel tools: email it, put it on a shared drive, run it off your laptop.\n\nWho This Is For\nAnalysts in environments where you need to work with data locally without remote databases\nTeams with no server access where IT says deployment will take months or isn't possible at all.\nFast prototyping scenarios where you need answers this week, not next quarter.\n\nReal-World Usage\n\nOne client's finance team uses this exact single-file app to process weekly reports. Multiple CSVs previously handled through Excel pivot tables and VLOOKUPs now run through a multi-step AI process with validation built in. Output: clean CSV ready for final Excel pivot analysis. No server. No deployment approvals. Just the HTML file on their shared drive.\n\nWhat I Built vs. The Original SQL Rooms\n\nSQL Rooms (sqlrooms.org) provides the foundation: DuckDB in browser + AI text-to-SQL interface.\n\nMy customizations:\nExpanded file type support: CSV, TSV, pipe-delimited, plus intelligent delimiter detection\nExport capabilities: Export individual tables or entire databases. Build intermediate work products, export them, share them.\nIterative debugging with AI: API errors and query failures now get passed back to the AI agent with context for self-correction across multiple steps. Added guidance for typical errors and debugging protocols. If it stops mid-process, prompt it to continue debugging.\nAI-driven schema detection: The AI examines schema before running queries, reducing manual setup.\nDatabase management: Clear all tables and start fresh when needed.\n\nExample Workflow: Building Transaction and Customer Datamarts\n\nImported 1M customer records and 10M transaction records (650MB CSV) into browser-based DuckDB. Had AI create ~25 feature from raw transactions, put that into a transactions datamart of one row per customer, merge with the customer data to create customer datamart and generate a segmented profile report with 40+ KPIs and charts. Natural language queries throughout.\n\nTwo Deployment Options\nServer-based deployment - For larger teams operating within corporate VPNs. Deploy once, multiple users access via internal URL. Suitable when you have infrastructure but need fast AI-powered querying without building custom applications\nNon-server deployment - For environments where server deployment isn't approved or feasible. Download the 3.5MB HTML file. Run it locally. Share via email or shared drives. No installation, no backend, no IT tickets.\n\nDecision criteria: Small team with no IT support or data privacy requirements? Non-server. Multi-department deployment with existing infrastructure and data governance protocols? Server-based.\n\nReal-World Caveats\n\nThis demonstration uses clean synthetic data to show tool capability and workflow concept. Real-world analytics always requires iterations, data cleaning, validation, and error handling. No tool - AI or otherwise - delivers production-ready analysis in one click. This app provides the framework and capability. You bring domain knowledge, validation discipline, and iterative refinement. That's how live analytics works.\n\nTry It\n\nUse the deployed version or download the bundled single-file app.\n\nGet started in 3 steps:\nGet a free Gemini API key from aistudio.google.com (takes 2 minutes)\nLoad the demo datasets or upload your own files\nRun your first natural language query\n\nTest datasets available: Customer and transaction files (1M + 10M records) on my Google Drive (link in Resources section below).\n\nSample Prompts and Results\n\nPrompt 1: Build Transaction Datamart\n\nUse the transaction table to build a transactions datamart with one record per customer. Should be a NEW TABLE.\n\nIn the transaction table (10M records):\n1001 = cash transactions\n1002 = retail sales\nuse the AMOUNT field\n\nFocus on customer transaction behavior, summarizing cash vs. retail, averages, counts, values, and other derived features. Give thought to creation of these derived transaction variables so that they are insightful and useful for upcoming deep dive analysis and model build.\n\nShare variables that you created, categorized by intuitive categories along with pseudo code.\n\nMaximum new variables to be added: around 25. Use multiple queries as needed.\n\nGo ahead and create the trans datamart table.\n\nTransaction Datamart\n\nTransaction Variables\n\nPrompt 2: Customer Profile Report\n\nNow next step: There is a customer file. Create a new customer datamart by merging the customer data with this transaction datamart.\n\nThen generate a customer profile summary report based on this customer datamart, providing an overview of customer characteristics. Break it down by housing variables to offer insights at both the overall and segmented levels.\n\nIncorporate as many relevant features from the datamart as feasible to create a clear snapshot of customer profiles. Make sure that the profile report has at least 40 KPIs properly grouped into categories.\n\nShare in a nicely formatted table format - vertical format with housing segment in columns and the KPIs in rows.\n\nAlso share 3 insightful charts based on the final profile summary that you create.\n\nCustomer Profile Report\n\nCustomize It\n\nSource code: Hit 'Docs' on the app site (www.tigzig.com/sql-rooms)\n\nWork with your AI Coder (Cursor, Claude Code, Gemini CLI) to customize for your specific workflows. The GitHub repo includes architecture documentation explaining modifications and deployment details for quick reuse.\n\nHow It Compares to My Other Database AI Tools\n\nThis is one of 10 open-source Database AI micro-apps I've built, each serving different deployment scenarios:\nDATS-4 www.tigzig.com/analyzer - React UI connecting to remote databases (PostgreSQL, MySQL). Handles situations where you have database infrastructure and need team-wide access. If uploading files, creates temporary tables on remote database.\nChatGPT + Database connectors: For rapid deployment where ChatGPT interface is acceptable. Direct database connections with minimal setup.\nFlowise AI solutions: Provides both backend and frontend. Native Flowise interface connects to databases for teams already using Flowise workflows.\nThis tool (DABX-1): For local execution, data privacy, and zero-infrastructure deployment. When remote databases aren't an option or you need offline capability.\n\nThese tools are complementary. Use based on your infrastructure constraints, privacy requirements, and deployment timelines.\n\nAll tools available at www.tigzig.com\n\nTechnical Note: Is It 100% Local?\nFile data: Remains in browser's DuckDB instance. Never uploaded.\nLLM receives: Schema, sample rows, and query results for generating SQL. Not your full dataset.\nAPI calls: Go directly from browser to LLM (Gemini, OpenAI, Claude).\nCan it be 100% offline? Yes. The original SQL Rooms repo supports Ollama for fully offline LLM use.\n\nMemory and Performance Limits\n\nHandles multiple files of a few hundred MBs to GB+ with ease. For larger volumes, performance depends on your device memory. DuckDB-WASM defaults can be tuned if your hardware supports it.\n\nResources\nSource code and docs: github.com/amararun/sql-rooms-tigzig-final\nDocs also accessible via app - 'Docs' tab\nTest datasets (Google Drive): 1M customer + 10M transaction records\nOriginal SQL Rooms project (credit): sqlrooms.org by Ilya Boyandin\n\n\nRelated\n\nTools: BRIQ, DATS-4 Database AI Suite, DUCKIT - CSV to DuckDB\n\nExplore: Database AI Apps hub, xlwings Lite Practice Lab, Browse all apps\n\nMore posts: Database AI, built for day-to-day work. Five categories, ten micro apps. Live, open source, free., BRIQ App: DuckDB AI in Browser - 500MB Files, 4M+ Records, No Database Setup, Run a Full AI Database App as a Single HTML File. No Server. No Remote DB., Chat, Query, and Transform Multi-GB Files - In Natural Language, Right in Your Browser with DuckDB."
},
{
"slug": "sebi-mf-categorisation-changes-jul2026",
"title": "SEBI Has Revised Mutual Fund Categorisation and Introduced a New Fund Type. The Major Changes, and What They Mean for MFPRO Users and Analysts.",
"tags": [
"mutual-funds"
],
"content": "SEBI Has Revised Mutual Fund Categorisation and Introduced a New Fund Type. The Major Changes, and What They Mean for MFPRO Users and Analysts.\n\nSEBI has modified mutual fund categorization and introduced new fund type...sharing the major changes, what it means for Tigzig MF API users and analysts.\n\nThe circular came on 26 February 2026. Existing schemes have six months to comply, so by late August every fund house will have re-filed its schemes under the new categories.\n\nThe circular: Categorization and Rationalization of Mutual Fund Schemes\n\nThe circular explains the changes very nicely .. key points below:-\n\nKey changes in the circular\nA brand new category called Life Cycle Funds. These are target date funds, e.g. Life Cycle Fund 2041. They start equity heavy and automatically shift toward debt as the maturity year approaches. Zerodha has already launched the first two.\nSolution Oriented schemes are discontinued. Children's funds and retirement funds stop taking new subscriptions and will be merged into other schemes.\nSectoral and Thematic are now two separate categories, with portfolio overlap limits between them.\nMost debt categories are renamed. Short Duration becomes Short Term, Dynamic Bond becomes Dynamic Term Fund, and so on across the duration ladder.\nFund names must match the category. A large cap fund has to be called Large Cap Fund, and words that hint at returns are not allowed in names. As I understand it using terms like 'Bluechip' and 'Opportunities' etc. in scheme name not allowed. Expect a wave of fund renames over the next month.\n\nWhat the transition looks like in the AMFI data\n\nWe watch AMFI's daily NAV files closely at TigZig MFPRO, and the transition is already visible in the raw data. New fund filings started carrying the new category headers from 17 July. One fund launched as Multi Asset Allocation and was re-filed as a Life Cycle Fund within two weeks.\n\nAnalysts would need to be mindful that old and new category / sub-category names will coexist in the data for a while...as that can cause issues in analysis pipelines.\n\nWhat this means for MFPRO users (Tool and API)\nThe top level category view moves to four clean buckets: Equity, Debt, Hybrid and Others. Life cycle funds will sit inside 'Others' (categorygroupclean)\nThe exact category text as filed by AMFI stays on every scheme, so you can always filter on the raw category or sub-category. Search is available on the tool as well as on API now\nScheme names will follow the latest AMFI filing. Until now we showed the name a scheme was first filed under. With the rename wave coming, we are switching to the current name, so what you see matches what the fund house calls the fund today.\n\nThe change would be carried over the next few days and data dictionary and docs updated once done.\n\nThe category and sub-category fields\n\nFull data dictionary on the docs site on MFPRO tigzig.com/mfpro. Some quick points regarding the category and sub-category fields. The counts below are as they stand today and are moving as new filings come in.\n\ncategory\n\nRaw category text from the AMFI header. About 65 distinct values accumulated over time. Modern: 'Equity Scheme - Large Cap Fund'. Old-style: single words like 'Income'. Prefer the clean fields below.\n\ncategory_group\n\nRaw field from AMFI. About 21 distinct values. For practical use, use the categorygroupclean field below.\n\nsub-category\n\nAbout 52 distinct values. e.g. 'Large Cap Fund', 'Liquid Fund'...BLANK for ambiguous old-style schemes by design; those rows will not appear in sub-category filters.\n\ncategorygroupclean\n\nRECOMMENDED category field. Now exactly 4 values: Equity Scheme, Debt Scheme, Hybrid Scheme, Other Scheme. The update is done, and it went to four rather than the five I first expected, because Fund of Funds moved as well as Solution Oriented. Both of those now sit inside Other Scheme, and their schemes stay fully findable through the raw category group and the sub-category. Old-style names are mapped in deterministically (Income -> Debt, Growth -> Equity, Balanced -> Hybrid, etc.).\n\nQuick links\nMF Analytics Tool: tigzig.com/mfpro. Go to Docs for Data Dictionary, Definitions, Methodologies etc.\nMF API Hub: tigzig.com/apis/mf-nav\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Mutual fund analytics hub, MF NAV API docs, API + MCP hub\n\nMore posts: MFPRO Mutual Fund API Update. Three Things Your Calls Asked For, Now Built., Data Dictionary & Docs Live on MFPRO. Every Field Explained, 21-Metric Methodology, and How the India MF Universe Actually Looks., One Line of User Feedback, Fixed Within a Day. MFPRO Search Upgraded + Complete AMFI NAV History, 37M Rows, 2006 Onward., 227 Downloads, 41GB in 10 Days. India MF Complete NAV Database - 38K+ Schemes, 37M+ Records, One 167MB Parquet. Free."
},
{
"slug": "sec-financials-three-ways-xbrl-edgartools-jul2026",
"title": "Pulling Company Financials Straight From SEC Filings. Free, No Vendor, No API Key. Three Routes Tested Head to Head.",
"tags": [
"portfolio-quants",
"ai-coders"
],
"content": "Pulling Company Financials Straight From SEC Filings. Free, No Vendor, No API Key. Three Routes Tested Head to Head.\n\nWanted to analyze the cash flows for Mag 7. Spent a session in Claude Cowork pulling company financials straight from SEC filings. The setup that worked: edgartools Python library + edgartools Skills + SEC EDGAR API, run by an AI cowork app, Claude Cowork or ChatGPT Work.\n\nClaude Cowork can pip install it on its sandbox and run it for you .. and if you are working locally with Claude Code then you (or cc) would need to install it on your machine.\n\nWanted to recategorize cash flow line items into my custom buckets. Tried out three routes on the same six quarters and seven companies, the SEC XBRL API directly, edgartools, and yfinance. All three agreed on the numbers. The difference is in coverage and traceability.\n\nEdgarTools is the first call for this. Comes packaged with features like statement to pull a full statement in one shot with indent level and parent, tag standardization layer that can be customized, handles de-cumulation, full text search across EDGAR, 13F institutional holdings, TTM calculations and segment facts and more.\n\nGreat value add is skills which make it easy to work with an AI Agent (which I am using) and an MCP server (which I have not tried yet).\n\nThe raw SEC API stays useful for a short list of specific jobs and as a second path for number cross-validations. That checking earned its keep, the one number that disagreed turned out to be my own error, a restatement I had handled wrong.\n\nedgartools is by Dwight Gunning, MIT licensed, past 2,000 stars, and it ships Claude skills and an MCP server: https://github.com/dgunning/edgartools\n\nRead the full write-up\n\nCovers the basics of SEC XBRL filings as well as a comparison of the approaches, in a FAQ manner and plain language. Every script, dataset and filing behind the numbers is linked from the page itself, so you can read the code that produced any figure.\n\nPulling company financials three different ways - the full write-up\n\nWhat is in it:\n01 Concept, tag and fact\n02 The four routes, as slices of a cube\n03 Why not use frames for everything\n04 The taxonomy question\n05 Saving filings to disk\n06 R-pages, and what they look like\n07 Standardisation\n08 In-house versus outside money\n09 edgartools, and what a statement is\n10 Duration versus instant\n11 yfinance, and the reconciliation\n12 Which one to use with an AI agent\n13 What else edgartools does\n14 The skills, and reading them properly\n15 Where to find it\n16 Points to watch\n17 Index of scripts and files\n\n\nRelated\n\nTools: QRep - Security Reports, Quants Agent, TREMOR - Macro Stress Signals\n\nExplore: Claude for Analytics hub, API + MCP hub, Analysis hub\n\nMore posts: Jamie Dimon Uses Claude Code to Generate Dashboards. Are You? You Need an AI Worker or AI Coder, Not an AI Chatter., AI Is Not Just Chat. Cowork Apps Are Where the Action Happens., Claude Cowork and Claude Code Have Quietly Become My Universal Front End. More Than Excel, Jupyter or Colab., Every Tigzig MCP is now Streamable HTTP. Requesting existing users from LM Studio, ChatGPT, Claude etc to make a one time change if you have not already done so"
},
{
"slug": "security-checklist-web-apps-71-items",
"title": "Security Checklist for Web Apps - 71 Items",
"tags": [
"security",
"infrastructure"
],
"content": "Security Checklist for Web Apps - 71 Items...all practical stuff...\n\nSecurity Checklist\n\nMy endpoints got hammered recently. I deserved it. My AI Coder (Claude Code now) flags security issues on every audit. Some I fix.. some I plan to do 'in a bit'.. but never come around to it...\n\nClient apps: always hardened with OAuth, rate limits, API logs - that's mandatory. But my 30+ public apps... I tend to get negligent.\n\nNow I've setup a proper checklist.... 71 items across React, FastAPI, Postgres, DuckDB, Cloudflare, MCP servers, Auth & VPS security... all detected in my apps... with the fixes + practical issues I faced when implementing the fixes. Each item in plain English with a basic code fix for reference.\n\nThere is nothing like a 100% checklist ...would be updating this as I go along\n\nNow every app - my AI Coder has to run it past this checklist before release. No exceptions...at least that's the plan.\n\nIf you come from a data science background like me - this stuff is new. Otherwise, IT world has been doing this for decades. But for analysts and data scientists now building tools... whether freelance or inside a company... very very important stuff..\n\nFull checklist on site with a copy-to-markdown button. Paste it to your AI coder - it will explain, execute, customize for you..\ntigzig.com/security\n\nEarlier Infra Guides\nPart 1: AI Coder\nPart 2: Deployment & Hosting\nPart 3: Security (updated) this one... tigzig.com/security\nPart 4: CORS\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Security checklist, MCP servers and agents, Database AI apps\n\nMore posts: 2026 Infra Guide for AI Tool Builders - Part 3: The 18 Common Security Mistakes and How to Fix Them, 2026 Infra Guide for AI Tool Builders - Part 5: Perimeter Security on Cloudflare (Free Plan), My Public MCP Server Got Hammered - Security Lessons from a Bot Attack, Your API Key Is Visible in the Browser. Even if you put it as Vercel's 'secret' backend env variable., Server Meltdown: How Bots Crashed My AI Tools and What I Did About It"
},
{
"slug": "security-performance-report-for-investors-ai-quant-agent-live-open-source-free",
"title": "Security Performance Report for Investors. AI Quant Agent. Live. Open Source. Free.",
"tags": [
"portfolio-analytics"
],
"content": "Security Performance Report for Investors. AI Quant Agent. Live. Open Source. Free.\n\nAnalysis generated by the Quants Agent, one of my open-source tools. The agent was instructed to run a full SPR on these securities.\n\nKey outputs from the report:\n\nRisk/Reward Visuals: CAGR vs. Max Drawdown, plus Sharpe/Sortino ratios.\n\nPerformance Tables: Full metrics including Return, Drawdown Durations, and my Anxiety Index.\n\nReturns: Cumulative returns, interactive charts, monthly returns\n\nDrawdown Analysis: Details on the 7 worst drawdown periods for each security.\n\nData Downloads: Processed price data and cumulative returns available as raw CSVs.\n\nHow to run your own analysis:\nGo to: www.tigzig.com (Path: Quants & Portfolio Apps -> Quants Agent)\nInstruct Agent: Just type 'run security performance report for X, Y, Z' symbols.\nSymbols: Use Yahoo Finance tickers or let agent search for it.\n\nMethodology: Full documentation comparing this tool's methodology, validations, reconciliations. No black boxes.\n\nOpen Source: All frontend (React) and backend (FastAPI/ MCP Servers) code is public. Use the live tool or customize the source for your own requirements.\n\n\nRelated\n\nTools: Quants Agent, QRep - Security Reports, Quants Suite\n\nExplore: Security Performance Report API, Quant Apps hub, API/MCP hub\n\nMore posts: Quants Agent: Now with LLM Choices for Technical Analysis Reports, S&P 500 Drawdown Analysis with QREP, Rolling Returns: Why CAGR Alone Can Mislead You (And What To Use Instead)"
},
{
"slug": "self-hosting-infrastructure-ai-tool-builders-2026-part-1-ai-coder",
"title": "2026 Infra Guide for AI Tool Builders - Part 1: AI Coder",
"tags": [
"ai-coders",
"infrastructure"
],
"content": "2026 Infra Guide for AI Tool Builders - Part 1: AI Coder\n\nClaude Code\n\nI build AI and automation tools for small businesses. \n\nBeen doing this for 2 years now. Hard lessons after defending against bot attacks, having API keys exposed and credits vanish, AI agents running wild on my bill, users watching a file operation take 10-15 mins instead of under a minute, automations timing out, API failures.\n\nWhen things break at 2am - API calls fail, tables get deleted, user data vanishes, AI agents devour credits, automated reports don't tally - there's no IT team. The buck stops with me. Needs to be fixed. Fast.\n\nThis series covers the infra setups where rubber hits the road - AI coders, VPS, deployments, security, monitoring, auth, build vs buy. Sharing what's working for me after those hard lessons.\n\nThis setup runs tigzig.com's 30+ production tools and client projects. All live. All open source.\n\nThe single biggest factor has been AI Coders. Without it, none of this would exist.\n\nI now use Claude Code for everything. Earlier Cursor. The whole of tigzig.com and numerous client tools all built with it.\n\nWhat Claude Code Actually Does for Me (Beyond Writing JavaScript & Python)\nBuild full-stack apps - React frontends, FastAPI backends, PHP integrations. Not snippets - complete deployable systems.\nDeploy directly - create projects on Vercel & Coolify, set environment variables, mount volumes, debug build failures.\nManage databases - setup new DBs, configure role-based access tied to Auth, handle migrations between providers (Supabase to Neon), manage connection pools.\nSetup and manage Auth - across Auth0, Clerk, Supabase, Neon depending on what client uses.\nHandle DNS on Cloudflare - setup subdomains, attach to deployments, create Cloudflare workers, manage cache rules.\nDebug production issues - SSH into servers, inspect docker container logs, diagnose SSL/connection errors, trace API failures.\nRun Git operations - setup repos, manage commits and PRs, handle branch management, review diffs.\nSetup API monitoring and logging - centralized logging tracking API calls across apps.\nRun security audits - review code for vulnerabilities, manage API keys, implement IP hashing for privacy.\nUpdate legal pages - website terms and privacy based on actual codebase to ensure transparency and compliance.\nManage server resources - check disk usage, clean up old docker containers and images, optimize storage.\nBuild file processing tools - data converters, web scrapers, data processors, PDF extraction.\nResearch and offline analysis - web search for technical solutions, data analysis with reports shared as markdown or PDF.\nCreate internal tools - API monitors, blog managers, markdown to PDF converters.\nBuild Windows desktop utilities with C# / .NET 8.\n\nAnd lot more. These are just top of mind.\n\nHow This Actually Works\n\nThis isn't where you can blow hot air into the prompt and an app appears magically.\n\nI come from a data science background. Most of this infrastructure work - FastAPI backends, Cloudflare workers, serverless architecture - was new to me. Claude Code didn't just build it, it trained me. Explained concepts in simple terms. Helped me understand trade-offs.\n\nArchitecture I still manage. But I brainstorm with Claude Code - interrogating it, asking questions, evaluating options. For some of my tools, I spent nearly a day and a half in discussions. Just discussions. Multiple plans and markdown documents without a single line of code written. Because I need to understand it, evaluate it, make sensible calls.\n\nI have a separate post on this - how I was working for two days doing only brainstorming and discussions before writing any code: https://www.tigzig.com/post/biggest-lesson-2025-ai-writes-better-code-when-you-dont-let-it-code\n\nThe grind doesn't go away. API call failures. JSON parsing errors. Validation work. Stress testing. Edge cases. Multiple user scenarios. Hours of debugging. Claude Code accelerates execution and helps with debugging, but you still put in the work. It doesn't replace judgment and grind.\n\nMulti-Project Management\n\nI run 30+ tools on my site and and ongoing stream of client projects. Work on multiple (related) apps at a time on VS Code. Multiple Claude Code sessions running per project. Sometimes using branches, sometimes git worktree, sometimes just parallel sessions - depends on what kind of work is happening.\n\nTop Tips\nSkip MCP where CLI/API exists - saves tokens. Vercel, GitHub, Auth0, Supabase, Cloudflare all have CLI/API. And Claude Code loves it.\nUse global and local CLAUDE.md for how you want things done - keep it tight and focused.\nRun multiple sessions in parallel, use branches and git worktrees for max efficiency.\nAnything you're doing manually - take a step back and ask Claude Code if it can do it. Most likely answer \"Yes I can do it. I can do so and so. Shall I implement it for you?\"\nContext is everything for infrastructure work. Feed it your current server setup, existing auth configuration, database schemas, deployment details. Better context = better output.\nCheck out this recent post from Boris Cherny, creator of Claude Code, on the setup he uses: https://x.com/bcherny/status/2007179832300581177\n\nThe Biggest Mistake with AI Coders\n\nNot understanding architecture. Just asking it to \"go build something.\"\n\nThere's nothing called \"go build\" and it appears. You have to go deep into the architecture, understand it, make sensible calls, test it. There are hours and hours of API call failures, JSON errors, validation work, stress testing with multiple users, different edge cases. It's a whole lot of grind. Claude Code does the coding, but you still need to validate, test properly, make it work.\n\nI use the same 6 rules I documented in my earlier post on working with AI coders: Share context, Ask for alternatives, Tell what you want, Iterate, Validate, The grind doesn't go away: https://www.tigzig.com/post/ai-coders-give-you-the-edge-the-6-rules-i-follow-when-working-with-ai-coders\n\nCosts and Value\n\nI use the $200/month tier. Multiple parallel sessions without thinking about it. The multiplier effect justifies it completely. You can start at $20 and scale as needed.\n\nStarting out with AI Coders? Google Antigravity is easiest entry point. Gemini CLI has generous free tier and great for many tasks as you start off but stumbles for many complex pieces.\n\nGreat guides and tutorials on YouTube, Medium and others where experienced developers have been generous to share how they use Claude Code for various tasks. \n\nAs for my client codes, deployments, and access to my servers, databases and DNS - Nobody touches it except Claude Code.\n\nI work alone. That's not my AI coder - that's my full dev team.\n\nAnthropic doesn't have a moat, they have a gorge. They may not know it but they've got me by the throat - whatever they charge, I will swipe.\n\nInfra Guide Series\n\nThe full \"Infra Guide for AI Tool Builders\" series:\nPart 1: AI Coder - Claude Code across the full build-to-deploy chain\nPart 2: Deployment & Hosting - Hetzner, Coolify, Vercel, Cloudflare\nPart 3: The 18 Common Security Mistakes (and Fixes) - production security hardening\nPart 4: Free CORS Proxy - Cloudflare Worker Code - copy-paste Worker to bypass CORS\nPart 5: Perimeter Security on Cloudflare (Free Plan) - edge defense, WAF, rate limits, honeypots\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Browse the apps, Market tools hub, Security checklist\n\nMore posts: 2026 Infra Guide for AI Tool Builders - Part 2: Deployment & Hosting, 2026 Infra Guide for AI Tool Builders - Part 3: The 18 Common Security Mistakes and How to Fix Them, AI Coders give you the edge.The 6 Rules I Follow When Working with AI Coders., Biggest lesson from 2025: AI writes better code when you don't let it code, Claude Code Was Getting Dumber. Stopped It from Auto-Spawning Sub-Agents. Performance Came Back Instantly."
},
{
"slug": "sif-august-2026-biggest-month-sep2026",
"title": "August Was the Biggest Month Indian SIFs Have Had. Net Assets Rs 31,175 Crore, Net Inflows Rs 7,699 Crore.",
"tags": [
"mutual-funds",
"database-ai"
],
"content": "August Was the Biggest Month Indian SIFs Have Had. Net Assets Rs 31,175 Crore, Net Inflows Rs 7,699 Crore.\n\nSIF data at sif.tigzig.com is current to 11 September, refreshed twice a day.\n\nAugust was the biggest month the category has had. Net assets went from Rs 23,177 crore to Rs 31,175 crore, net inflows were Rs 7,699 crore, which is more than May and June put together, and folios went from 94,000 to over 125,000.\n\nThe fastest mover is Equity Ex-Top 100 Long-Short, up about 66% from Rs 3,790 crore to Rs 6,289 crore, and it now has seven schemes across seven fund houses. Hybrid Long-Short is still where most of the money sits, about 63% of the total.\n\nTwo of the seven strategies SEBI permits still have no funds at all, both of them the debt ones.\n\nFree, no signup, CSV and Parquet.\n\nThe mutual fund side is a separate thing from the SIF piece\n\nThe live tool is at tigzig.com/mfpro. How that database is built and validated, and the data dictionary, are in the Docs section.\n\nAPI docs are at tigzig.com/apis/mf-nav. You can pull NAV one scheme at a time, batch up to 50 schemes in a single call, or take the whole history as one file. Or download manually from the app, under Data and API.\n\nSIF is kept out of that pipeline for now, and the identifier is the main reason. A SIF scheme code is text with a prefix, SIF-120, while a mutual fund scheme code is a plain number like 119551, so the two cannot be stacked into one column as they are. There is more than one way to do it and it will happen at some point, once this collection has run on its own for a while longer.\n\nLinks\nSIF data, free download in CSV and Parquet: sif.tigzig.com\nMFPRO, the live mutual fund tool: tigzig.com/mfpro\nMF NAV API docs: tigzig.com/apis/mf-nav\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, MF File Converter, DUCKIT - CSV to DuckDB\n\nExplore: MF NAV API docs, Mutual funds hub, API and MCP catalog\n\nMore posts: SIF Is India's Newest Fund Category and It Nearly Doubled in Three Months. Every NAV AMFI Has Published Is Now Downloadable., Heavy User of the Mutual Fund NAV API? Run Your Own NAV Service Inside Your Own Setup, India MF Tool Builders: You Do Not Need a Thousand API Calls a Day. The Whole NAV Dataset Is One File., MF NAV in the Wild. A Review of the API Logs, and How Widely the Mutual Fund NAV Data Is Being Pulled., MFPRO Mutual Fund API: Search Now Filters Across 12 Fields on a Scheme"
},
{
"slug": "sif-data-release-nav-scheme-master-aum-aug2026",
"title": "SIF Data Is Live. The Full Published NAV History, the Scheme Master and the Quarterly AUM.",
"tags": [
"mutual-funds",
"infrastructure"
],
"content": "SIF Data Is Live. The Full Published NAV History, the Scheme Master and the Quarterly AUM.\n\nSIF data is live at sif.tigzig.com. It carries every NAV that AMFI has published for Indian Specialized Investment Funds since the category began, along with the scheme master and the quarterly average AUM. Plain CSV, updated twice a day, no sign-up. This post covers what is in the files, what to watch for when you use them, and what has to be worked out before this goes into MFPRO.\n\nQuick Info — IP Address Blocks (CIDR Notation)
\nIn IP addressing, ranges are expressed as \"blocks\" using CIDR notation:
\nA /32 is a single IP (1 address)
\nA /24 is a block of 256 addresses (last number varies: 185.177.72.0 to 185.177.72.255)
\nA /16 is a block of 65,536 addresses (last two numbers vary)
\nSo in one of my cases I put up a /24 block (256 IPs). In this case the attacker owned two /24 blocks (512 IPs total), so blocking both /24s covered their entire operation. You're not blocking the whole internet — just a specific slice you've identified as malicious. But again, as mentioned earlier, this level of blocking is more of an exception than the norm.
\n\n
\n\nWhat you get\n\nSix files, updated twice a day. It is raw filed data and counts, so everything derived stays out and you get what AMFI filed along with our copy of it.\n\nFunds and scheme variants are two different counts\n\nAt fund level AMFI counts 27. Its SIF Monthly Report for June 2026 shows 27 schemes as on 30 June 2026, across 75,032 folios, with net assets under management of Rs 17,858 crore and average net AUM for the month of Rs 16,050 crore. Net AUM as on 30 June 2026 divided by folios on the same date works out to about Rs 23.8 lakh per folio. That report is published at strategy level only and carries no scheme code, so there is no way to join it to anything per scheme.\n\nThe files here hold scheme variants, which is the level AMFI files NAVs at. A fund offering a Direct and a Regular plan, each with a Growth and an IDCW option, appears as four separate rows, each carrying its own code, its own NAV and its own AUM, so add the variants up if you want the figure for the fund as a whole. As of early August 2026 that comes to 111 variants from 17 fund houses, and the first NAV is 8 October 2025, so the whole history is under ten months long. Of those, 63 are open ended and 48 are interval fund schemes.\n\nThe six files\nsif_nav.csv. Every NAV, every scheme, every date. About 10,000 rows across 200+ trading dates.\nsif_meta.csv. The scheme master, one row per scheme variant.\nsif_aaum.csv. Quarterly average AUM per scheme variant, three quarters so far.\nsummary.json. Precomputed counts and series, if you want the shape of the category without reading the CSVs.\nmanifest.json. Row counts, byte sizes, SHA-256 per file, and generated_at.\nREADME.txt. Every column, the units and the conventions.\n\nThe scheme master is closer to a category master today\n\nIt carries the identifiers, the fund house, the SEBI category and strategy, the plan and the option, and the first and last dates a scheme has a NAV for, which is the same derivation MFPRO uses. What it does not carry yet is the metadata that sits in a scheme's offer document, things like the scheme launch date and the fund manager. I plan to add those as this goes along.\n\nThe AUM\n\nThe quarterly average AUM in sif_aaum.csv has gone from Rs 2,445 crore in October to December 2025, to Rs 8,124 crore in January to March 2026, to Rs 13,584 crore in April to June 2026. That is the average across the whole quarter, which is why it sits below the June monthly figure above.\n\nGetting the files\n\nTwo ways. Go to the download section on sif.tigzig.com and take the CSVs directly, they open in Excel and the whole set is under a megabyte. Or fetch them programmatically from stable paths that need no key and no parameters, with no rate limit to think about. The curl, DuckDB and pandas examples are on that same page.\n\nPoints to watch\nObservational status. No uptime commitment, and the shape of the files may still change while the pipeline is in beta. Keep your own copy if you are building on it.\nScope. This is every SIF NAV AMFI has published since the category began, as far as I can determine. It is not complete SIF data in some absolute sense.\nPublished as filed. Some of what AMFI has filed is unusual and nothing gets corrected. The oddities are listed on the dataset page, each one checked against the source file before it was written down.\nThe temp prefix. Any column named tempsomething exists only while SIF is a separate dataset and will not be there once it folds into the main scheme master. Code written against a temp_ column breaks on the day of the merge. source has no prefix, because it is half the merge key.\nupload_ts is IST. AMFI stamps its publication timestamp with a trailing Z and the value is IST. Anything computing publication lag from it is 5h30m out unless it converts.\nSame column names as the mutual fund dataset. Same names, same order, with SIF-only columns appended after the shared ones, so code that already parses the mutual fund scheme master will read these files too.\nisin and isin2 are one scheme. On an IDCW scheme, isin is the payout option and isin2 is the reinvestment option of the same scheme. A statement showing the reinvestment ISIN is the same fund.\nFreshness. Read generated_at in manifest.json. Each file carries its SHA-256 there as well.\n\nOpen issues before it goes into MFPRO\n\nMFPRO is the mutual fund app and API this will eventually join. A few things have to be worked out first.\n\nThe taxonomy is different\n\nSIF carries its own SEBI investment strategies, Equity Long-Short, Equity Ex-Top 100 Long-Short, Sector Rotation Long-Short, Active Asset Allocator Long-Short and Hybrid Long-Short. Every category filter, peer comparison and percentile rank in MFPRO would have to decide whether SIF sits inside the mutual fund universe or beside it. Ranking a long-short fund against Flexi Cap peers would mislead the reader, so that has to be settled before the two go together.\n\nThe benchmarks\n\nAll 23 indices in the system are long-only. A long-short fund measured against Nifty 50 TRI produces a beta and an alpha that read as meaningful without being so. I need a benchmark that fits the strategy and I do not have one yet.\n\nInterval funds\n\nClose to half the scheme variants are interval fund schemes. Their liquidity semantics are different from anything in MFPRO today, so the app has to learn them.\n\nThe scheme code\n\nMutual fund scheme codes are integers, like 118632. AMFI's SIF codes are strings, like \"SIF-120\". The public API returns scheme_code as a number, the parquet file every downloader consumes types it as a number, and the frontend types assume a number. Making that column a string to accommodate SIF would break the API contract, the file schema and the app at once, for everyone already using them.\n\nWhat happens today is that the numeric part, 120, is stored alongside a source column carrying 'SIF'. \"SIF-120\" is exactly reconstructable from the pair, so nothing is lost and nothing is invented. It is safe today because every mutual fund scheme code is 100,027 or higher, with none below 100,000 across 38,000+ schemes and twenty years of history, while a SIF code stripped of its prefix runs from 1 to 152. Nobody ever promised that gap would hold, so it is asserted in code from both directions.\n\nWhat to do permanently is open, and there are three options with no free one among them.\n\n| Option | What it does | What it costs |\n| --- | --- | --- |\n| A. Raw number plus a source column, running today | Stores 120 with source='SIF'. Lossless and reversible, and every value is one AMFI publishes. | Every join, filter and ad-hoc query has to carry source. Forgetting it produces no error. You get a SIF scheme mixed with a mutual fund scheme of the same number, and a plausible wrong answer. |\n| B. Offset into a reserved range | 900000 plus n. One column is unique on its own and no query can forget anything. | 900120 is a synthetic identifier matching nothing AMFI publishes, so anybody who sees it cannot look it up anywhere. It would mean inventing an identifier, which is the thing I said I would not do. |\n| C. Separate tables permanently | No collision risk at all, and no shared key to get wrong. | Every cross-dataset question becomes a UNION, and MFPRO's analytics have to know about both shapes forever. |\n\nI am on A with the guards, and it is provisional. What will decide it is whether the source-column discipline survives contact with real ad-hoc SQL, and I cannot know that in advance.\n\nWhen it merges\n\nNo date. It stays under observation while I watch how the pipeline behaves, and it moves once the issues above are worked out and the collection has been steady for a stretch.\n\nNuances and checks\n\nThe data is collected twice a day from AMFI's published sources. This is a new category with a short history, so a fair part of the work is checking that what we hold matches what AMFI filed. These are the same checks that run on the MFPRO mutual fund data.\n\nTwo sources for the same date\n\nEach run compares the published bulk file against the JSON API for the same date and records whether they agree.\n\nFive-day lookback\n\nEvery run re-reads the last five days, so anything AMFI changes after the fact gets picked up instead of being left at the value we first saw. Late arrivals and restatements go into separate tables, so you can tell which one you are looking at. On the first night one date held 103 schemes at first pull and 109 after the overnight re-pull, six schemes that AMFI published a day late, with no restatements.\n\nHoliday detection\n\nA day with missing NAVs can be a market holiday or a real gap, and the two have to be told apart. Indian liquid and overnight funds declare a NAV every calendar day, so a market holiday still carries several hundred mutual fund NAVs against roughly eight and a half thousand on a trading day. Good Friday had 584, Ambedkar Jayanti 690, and the Monday between them 8,603. The check works off a trading-day median.\n\nGap detection\n\nA scheme that stops reporting has to be visible. Comparing each scheme against a trading calendar does not work here, because the category is growing and the early dates carry very few schemes. So the check is simpler. If one scheme of a fund reports on a day, its siblings should report too, and anything missing gets flagged.\n\nColumn shifts and parsing\n\nAn outlier check will not catch a parsing bug. If a column shifts or a decimal gets truncated, the values still look plausible and nothing stands out as odd. So the check goes back to the archived original file and compares value by value. It runs as a separate script with its own simpler parse, so the same bug cannot show up on both sides.\n\nAUM reconciliation\n\nThe quarterly AUM is checked against AMFI's own published fund, brand and industry totals, stored as filed. Exact at every level across all three quarters so far.\n\nReconciled to AMFI's monthly report\n\nRolling our variants up to fund level should reproduce AMFI's own count. As at 30 June 2026 it does, strategy by strategy, with nothing adjusted to make it fit.\n\n| Strategy | Our variants | Our funds | AMFI |\n| --- | --- | --- | --- |\n| Equity Long-Short | 34 | 9 | 9 |\n| Hybrid Long-Short | 35 | 9 | 9 |\n| Equity Ex-Top 100 Long-Short | 17 | 5 | 5 |\n| Active Asset Allocator Long-Short | 10 | 3 | 3 |\n| Sector Rotation Long-Short | 4 | 1 | 1 |\n| Total | 100 | 27 | 27 |\n\nThis checks the strategy we derive from AMFI's category string, the fund-house attribution, and whether the collection is missing schemes, and a mismatch in any of them would show up here. The variant count is 100 as at 30 June and higher today, because more schemes have started reporting since. Anyone can open the same AMFI monthly report and count the same 27.\n\nThe one thing we infer\n\nSome fund houses name only the Direct variant of a scheme and leave its Regular twin unlabelled. Franklin files \"Sapphire Equity Long-Short SIF - Direct - Growth\" alongside a plain \"Sapphire Equity Long-Short SIF - Growth\", and the second one is the Regular plan. Six scheme variants are like this.\n\nWe work the plan out from the scheme's siblings, and the files record which values came from AMFI and which are ours, so you can filter on tempplanbasis = 'stated' and get only what was filed.\n\nThe test uses the fact that a Direct plan carries no distributor commission, so its NAV has to run ahead of the Regular plan of the same fund. That holds regardless of how the scheme was named. The first version of the test failed on four families AMFI had already labelled, and in each case the test was wrong rather than the data, so it had to be corrected first. It now runs across 22 labelled families and 2,224 shared days with no reversals. Five of the six inferred plans then confirmed. The sixth is an IDCW option the test cannot reach, and it is reported as not covered.\n\nResources\n\nFor people\nsif.tigzig.com. The dataset page. Live counts, the category as it has grown, the largest schemes, the data quality table, the oddities register, and the downloads.\nMFPRO. The mutual fund side this will eventually join. 38,000+ schemes, NAV history back to 2006, returns, risk, rolling windows and portfolio composition.\nAPI and MCP hub. Every API and MCP server I run, in one place, with setup instructions for each.\namfiindia.com. The official source. Refer to it for the authoritative data.\n\nFor your AI agents\n\nAI agents: the files are open and need no auth. Start with the SIF readme at https://sif.tigzig.com/README.txt, which carries the file list, every column, the units and the conventions. For the mutual fund API this will eventually join, the OpenAPI spec is at https://api.tigzig.com/mf/v1/openapi.json. For every other TIGZIG API, the site-level catalog is at https://api.tigzig.com/.well-known/api-catalog. Site index for agents: https://www.tigzig.com/llms.txt\n\nIf you find something that does not match, or a SIF scheme that is missing, use the Feedback button on the page (anonymous).\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Mutual fund analytics hub, MF NAV API docs, API + MCP hub\n\nMore posts: Is SIF Analytics Coming to MFPRO? On the List, No Date., MFPRO Scheme Master: Every India Mutual Fund Scheme in One Small File, Free, SEBI Has Revised Mutual Fund Categorisation and Introduced a New Fund Type. The Major Changes, and What They Mean for MFPRO Users and Analysts., Data Dictionary & Docs Live on MFPRO. Every Field Explained, 21-Metric Methodology, and How the India MF Universe Actually Looks."
},
{
"slug": "sif-data-universe-download-sep2026",
"title": "SIF Is India's Newest Fund Category and It Nearly Doubled in Three Months. Every NAV AMFI Has Published Is Now Downloadable.",
"tags": [
"mutual-funds",
"database-ai"
],
"content": "SIF Is India's Newest Fund Category and It Nearly Doubled in Three Months. Every NAV AMFI Has Published Is Now Downloadable.\n\nSIF (Specialized Investment Funds) is the newest fund category in India and growing fast, nearly doubling in three months to an AUM of Rs 23.2K crore at the end of July from Rs 12.3K crore at the end of April. For analysts wanting to dig into it, every SIF NAV that AMFI has published is now downloadable in full at sif.tigzig.com. CSV and parquet, refreshed twice a day.\n\nA few quick points on SIF naming and branding\n\nYou will not recognise any of the 17 fund house names. SEBI requires a SIF to carry a distinct brand name and logo, separate from the AMC's regular mutual fund business.\n\nICICI Prudential's SIF is called iSIF, SBI's is Magnum, Kotak's is Infinity.\n\nTo identify the sponsor, the SIF code will not help, it is a running number and nothing more. The ISIN will. Each SIF scheme's ISIN carries the sponsor's issuer prefix, the same first seven characters that sit on that AMC's regular mutual fund ISINs. INF174K is Kotak on a SIF scheme and on a Kotak mutual fund alike, INF109K is ICICI Prudential, INF200K is SBI.\n\nIf you are going to work with this data, build that mapping once and keep it. Seventeen rows, prefix to AMC, and you only redo it when a new fund house turns up. The scheme master on the site carries both ISINs, so you can build it straight off the download.\n\nThis is based on my observation. It held on all 17 when I checked it.\n\nFor you and your AI agents: sif.tigzig.com\n\nThe SEBI Master Circular\n\nFor the source, the SIF framework now sits inside the SEBI Master Circular for Mutual Funds as on 20 March 2026, Chapter 21, pages 296 to 311. It consolidates the original circular of 27 February 2025.\n\nA few important clauses:\n\n21.2.1 requires a SIF to have a distinct brand name and logo, separate from the AMC's regular mutual fund. The sponsor's name can appear in offer documents for five years from approval, the brand itself has to stand apart.\n\n21.3.2 permits only one investment strategy per category. That is why a fund house plus a strategy is a fund, and it is how AMFI counts them.\n\n21.4.1 sets the minimum at Rs 10 lakh per investor at PAN level, aggregated across all strategies of that SIF, and it does not count what the investor holds in the same AMC's regular schemes.\n\nCircular: sebi.gov.in/legal/master-circulars/mar-2026/master-circular-for-mutual-funds_100491.html\n\nA few more numbers\n\nSEBI permits seven investment strategies for a SIF and two of them have no funds at all. Both are the debt ones.\n\nThe money is concentrated in one strategy. Hybrid Long-Short holds about Rs 15,400 crore of the Rs 23,200 crore total, roughly two thirds of it, from 11 of the 30 schemes AMFI counted at the end of July.\n\nNet inflows in July alone were about Rs 4,900 crore, and the folio count was 94,447 at the end of the month.\n\nThe AUM figures come from AMFI's SIF Monthly Report, which is a separate file from the daily NAV feed and comes out within about ten working days of month end. July is the latest one published. The files on the site carry the daily NAV and the scheme master, they do not carry AUM.\n\nWhy SIF sits outside the main MF API, I wrote that up separately: tigzig.com/post/mfpro-sif-analytics-where-it-stands-aug2026\n\nsif.tigzig.com\n\nThe SIF data universe\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, MF File Converter, DUCKIT - CSV to DuckDB\n\nExplore: MF NAV API docs, API and MCP catalog, Market and macro tools\n\nMore posts: Is SIF Analytics Coming to MFPRO? On the List, No Date., SEBI Has Revised Mutual Fund Categorisation and Introduced a New Fund Type. The Major Changes, and What They Mean for MFPRO Users and Analysts., August Was a Heavy Month in the AMFI Data. 1,234 Schemes Renamed and 37 ISINs Swapped. If You Join on Scheme Code You Are Fine., An ISIN Checker for When a Mutual Fund Lookup Fails. And Why One Fund Carries Ten Identifiers., MFPRO Scheme Master: Every India Mutual Fund Scheme in One Small File, Free"
},
{
"slug": "sonnet-4-5-released-yesterday-now-live-on-dats-4-sql-agent-suite-solid-upgrade-but-more-4-2-than",
"title": "Sonnet 4.5. Released yesterday. Now live on DATS-4 SQL Agent Suite. Solid upgrade, but more 4.2 than 4.5.",
"tags": [
"database-ai",
"text-to-sql"
],
"content": "Sonnet 4.5. Released yesterday. Now live on DATS-4 SQL Agent Suite. Solid upgrade, but more 4.2 than 4.5.\n\n PDF Report Available at\n Database AI\n\nQuality: Sonnet 4 was already the best. 4.5 is clearly better, but not by a big margin.\n\nCost: Same as Sonnet 4.\n\nAssessment: Initial assessment based on quick tests (see below/ docs). More to follow as I use it live.\n\nScope: Reasoning based advanced analysis in multi-step multi-agent setup.\n\nQuality Scores\nClaude Sonnet 4.5: 115\nClaude Sonnet 4: 100\nGPT-5: 95\nQwen3 Max: 90\nDeepSeek R1: 85\nGemini 2.5 Flash: 75\n\nCost\n\nReasoning & planning cost for advanced analysis task in USD per 100 questions. Execution cost extra. See next paragraph / doc for details.\nClaude Sonnet 4.5: $6.5\nClaude Sonnet 4: $6.5\nGPT-5: $15.0\nQwen3 Max: $2.0\nDeepSeek R1: $2.25\nGemini 2.5 Flash: $1.75\n\nIn multi-step advanced analysis, execution is the biggest chunk\n\nIn DATS-4 I use GPT-4.1 for execution at about $10 per 100 questions. A single-step SQL agent is much cheaper (GPT-4.1 = $2 per 100, 4.1-mini = $0.50, 4o-mini = $0.25).\n\nMy recommendation: use single-step wherever possible\n\nIn my own live usage, approx. 80% of work runs through single-step agents. Multi-step - reserve for the tough analysis cases.\n\nTask Examples (tested live)\nCredit card data mart build (1M customers, 10M transactions, AWS RDS MySQL).\nWeighted score ranking of Indian issuers (RBI monthly file, Supabase).\n\nDATS-4\n\nFull-stack Database SQL multi-agent: text-to-SQL, Python charts, stats, instant Postgres DB, PDF outputs, 9 reasoning models.\n\nLive App (free)\n\nwww.tigzig.com\n\nAll credentials and API calls pass through my backend server. Free to test. For live use, deploy on your VPN.\n\nDocs and source: open, full code at app site.\n\nSonnet 4.5 Assessment\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, BRIQ, Database Query Tool\n\nExplore: Database AI Apps hub, API / MCP directory, Analysis hub\n\nMore posts: Qwen3 Max now live on DATS-4 SQL Agent Suite for Advanced Analysis Better than DeepSeek R1, closer to Claude Sonnet 4 - at a lower cost., Two models added to Database AI Suite this week: GPT-5.1 and KIMI 2 Thinking., Gemini 3 Pro on Database AI Suite - Claude 4.5 still leads"
},
{
"slug": "sp500-ath-cracking-underneath-oct2007-parallel",
"title": "S&P at All-Time High. Under the Hood, Things Are Cracking. Just Like October 2007.",
"tags": [
"portfolio-quants"
],
"content": "S&P at All-Time High. Under the Hood, Things Are Cracking. Just Like October 2007.\n\nPublished: 2026-04-18\n\nS&P at ATH while private credit at 2x 2007 subprime showing stress. Credit card and auto delinquencies above 2007 pre-crisis. Bank charge-offs higher than 2007. AI displacing workers at unprecedented pace inside a stress cycle. Tracking across FFIEC, FDIC, NY Fed, IEA, Challenger, Anthropic Economic Index.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/sp500-ath-cracking-underneath-oct2007-parallel\nMarkdown of the HTML page: send GET https://tigzig.com/post/sp500-ath-cracking-underneath-oct2007-parallel with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/DISBELIEFRALLY18APR2026.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nThe Disbelief Rally of 2026\n\nAmar Harolikar ACA\n\nDecision Sciences & Applied AI\n\nEx-Derivatives Trader\n\n18th April 2026\n\ntremor.tigzig.com\n\nIn Oct 2007, markets hit new highs while subprime was already cracking. Today the S&P is at all-time high. The cracks are in different places. The disbelief of reality is the same.\n\n2007 Repeating Itself - With an AI Wildcard\n\nNew High – Oct 2007\n\nNew Low – Mar 2009\n\nS&P 500\n\n~60% Drop\n\n2007 vs 2026\n\n | 2007 | 2026\n\nS&P at Highs | 1,576 (Oct) | 7,100+ (Apr)\n\nShiller CAPE | ~27x | ~40x (2nd time ever)\n\nUnemployment | 4.7% and rising | 4.3% - 35-mo drift from 3.4% trough\n\nHidden Losses | Subprime CDOs at par | Private credit BDCs at par\n\nHidden Leverage | ~$1.5T subprime | ~$2.7T private credit\n\nBank Losses | Rising in mortgages | Rising in Autos, cards, student loans.\n\nOil Shock / War | Oil rose, no supply cut | Hormuz 40-day closure. Brent $112+\n\nThe Variable 2007 Didn't Have\n\nAI displacement of labor. Entry-level hiring down & increasing layoffs of experienced workers. Computerization took decades. This is happening in one cycle, INSIDE another stress cycle.\n\nThe Disbelief Rally\n\nS&P 500 new highs preceded both episodes – the temporary euphoria from an immediate news event\n\nDisbelief Rally\n\nCrash Follows\n\n2007-2008\n\nDisbelief Rally - Same Setup?\n\n2025-2026\n\n“Fed Minutes Spark Record Stock Moves ” NPR | Oct-9-2007\n\n“S&P 500 notches first close above 7,100” CNBC | Apr-17-2026\n\nCAPE at ~40x - Second Time Ever\n\nShiller CAPE at ~40x as of Jan 2026 - a 135% premium to the long-term mean of ~17.3. Only the second time in 155 years the 40-point threshold has been breached - the first was the Dot-com peak in 2000.\n\nIn 2007, CAPE was ~27x before the crash. Today's reading is 50% higher than 2007's peak.\n\nDot-com Peak 44.2\n\nNow 38.9\n\nLong-term Mean ~17\n\n2007 Peak 27.5\n\nWhat is the Shiller CAPE? Developed by Nobel laureate Robert Shiller, it divides the S&P 500 price by the average of 10 years of inflation-adjusted earnings - smoothing out short-term profit cycles to show how expensive the market truly is. Readings above 25 have historically preceded below-average returns; above 40 has only occurred twice - both near major peaks.\n\nSource: Robert Shiller, Yale University / shillerdata.com via tremor.tigzig.com\n\nLabor Market - Same Creep\n\nU-1: Long-Term Unemployed\n\nNov 2007: 1.5%\n\nMar 2026: 1.8% ▲\n\nAlready exceeds 2007\n\nU-3: Headline Rate\n\nNov 2007: 4.7%\n\nMar 2026: 4.3% ▲\n\n35-mo drift from 3.4% trough\n\nU-6: Underemployment\n\nNov 2007: 8.5%\n\nMar 2026: 8.0% ≈\n\nMatching 2007 level\n\nAll three measures - long-term, headline, and underemployment - are rising, mirroring the pre-crisis pattern of 2007.\n\nSource: U.S. Bureau of Labor Statistics (BLS) / FRED via tremor.tigzig.com\n\nAI Displacement (1/2)\n\nEntry-Level Already In Recession\n\nThe headline 4.3% unemployment masks a recession-level shock at the bottom of the career ladder.\n\nRecent grad unemployment: 5.7% in Q4 2025, up from 5.3% in Q3. Underemployment at 42.5% - highest since 2020. (NY Fed)\n\nUnemployment rate for new college graduates could go 30%+(McDermott, CEO Service Now, CNBC,March’26)\n\nNew grads now 7% of Big Tech hires, down 50%+ from pre-pandemic. Under 6% at startups, down 30%+. (SignalFire)\n\n\"I'm worried that when this year's college graduates enter the workforce, we could see the highest unemployment rate among them in years - even without a recession.“. - Larry Fink, CEO BlackRock (2026 Infrastructure Summit, March 2026)\n\nSource: Federal Reserve Bank of New York, College Labor Market, Feb 2026\n\nAI Displacement (2/2): Senior Layoffs Accelerating\n\nAnnounced cuts up 44% YoY. AI-attributed layoffs projected at 9x 2025. The AI producers are the ones warning.\n\n2026 cuts across the board: Oracle 20-30K, UPS 30K, Amazon 16K, Dell 11K, Chevron 8K, Dow 4.5K, Block 4K, MS 2.5K, Atlassian 1.6K, Snap 1K. Challenger, Gray & Christmas; TrueUp; Layoffs.fyi\n\nAI can theoretically cover 94% of computer/math tasks but currently handles ~33%. 49% of jobs can now use AI in 25%+ of tasks - up from 36% in early 2025. Displacement so far is a fraction of what's feasible. Anthropic Economic Index - Massenkoff & McCrory, March 2026\n\nCompanies shifting budgets toward AI investments at the expense of jobs. Challenger CRO, Forbes, April 2026\n\n\"AI will disrupt 50% of entry-level white-collar jobs over 1-5 years, while also thinking we may have AI that is more capable than everyone in only 1-2 years.\"\n\nDario Amodei, CEO Anthropic, Jan 2026\n\nConsumer Credit - Cracks Widening\n\nPost-COVID forbearance unwind: delinquencies and charge-offs rising across credit cards, auto loans, and consumer portfolios - many now exceeding pre-COVID levels.\n\nBoth the banking system (FDIC) and broader consumer credit (NY Fed) confirm the same upward trajectory since forbearance expired.\n\nFDIC-Insured Banking System\n\nConsumer Credit\n\nSource: NY Fed Consumer Credit Panel / Equifax via tremor.tigzig.com\n\nSource: FDIC SDI via tremor.tigzig.com\n\nAuto Loans - Worse Than 2007\n\nDelinquencies now exceed GFC peak levels. A $1.67T market with cracks showing in both prime and subprime.\n\nNY Fed / Equifax (All Lenders)\n\nAuto 90+ DPD: 5.21% (Q4 2025) vs 3.05% (Q4 2007)\n\nTotal market: $1.67T outstanding (Q4 2025)\n\nS&P Global Auto ABS Pool ($176B)\n\nSubprime charge-offs: ~10%; prime: ~6%\n\nNot just subprime – prime is cracking too\n\nSource: S&P Global Auto Loan ABS Tracker – Feb 2026 Performance\n\nSource: NY Fed Consumer Credit Panel / Equifax via tremor.tigzig.com\n\nStudent Loans - The Pressure Nobody Had in 2007\n\nStudent loans were ~3% of consumer debt in 2007; now ~9%. $1.66T outstanding, 40M+ borrowers. A new drag on disposable income that didn't exist last cycle.\n\nShare of Consumer Debt\n\n$1.66T outstanding. 2007: ~3% of total debt. Now: 8.9% - tripled in 15 years\n\nThis debt didn't exist at scale in 2007\n\n90+ DPD Stock Rate\n\nQ4 2007: 7.5% → Q4 2025: 9.57%\n\nNew Delinquency Flow Rate\n\nQ4 2025: 16.35% - highest since 2009, exceeds GFC peak of ~12%\n\nForbearance ended Oct 2024; now competes with mortgage, auto, cards. No bankruptcy discharge; wages garnished w/o court order\n\nSource: NY Fed Consumer Credit Panel / Equifax via tremor.tigzig.com\n\nBanking System Charge-Offs Rising\n\nBanks are writing off more loans across all sectors.\n\nOverall NCO rate (all sectors): Q3 2007: 0.56% → Q4 2025: 0.61% (annualized) - already past 2007 pre-crisis levels\n\nConsumer NCO rate: Q4 2007: 2.74% → Q4 2025: 2.90% (annualized) - surpassed 2007 pre-crisis levels\n\nIt is not just consumer loans - the entire banking system is writing off more across C&I, CRE, and consumer portfolios combined\n\nSource: FDIC SDI via tremor.tigzig.com\n\nFDIC-Insured Banking System\n\nBank Lending to Private Credit & Non Banking Lenders\n\nNDFI lending at $1.57T (Q4 2025), grew 35% in a single year. Fastest-growing segment in banking - CAGR of 21.9% since 2010, nearly 3x the next-fastest segment. (Source: FFIEC Call Reports, | Analysis & interactive data: tremor.tigzig.com → US-NDFI)\n\nAnother $1T in undrawn commitments- total potential exposure: $2.6T. PE Funds category alone has 70% undrawn-to-outstanding ratio. (Source: FFIEC Call Reports, | Analysis & interactive data: tremor.tigzig.com → US-NDFI)\n\nLending default rates to reach 8%, approaching Covid peak levels. (Morgan Stanley, Mar’26)\n\nBank credit lines tightening - transmission confirmed. Banks, including JPMorgan are marking down private credit loan collateral and raising rates on back-leverage, prompting funds to swap out holdings (Bloomberg, CNBC, Capital Advisors Group)\n\n\"Banks fund private lenders through credit lines and could pose systemic liquidity risk to the banking sector under certain circumstances, including borrowers defaulting together... tail risk may be underappreciated.“ - Federal Reserve Bank of Boston, Current Policy Perspectives 25-8, May 2025\n\nBanks can't make risky loans post-2008. So they lend to the funds and Non Depository Financial Institutions that do.\n\nPrivate Credit - Ghosts of 2008\n\nQ1 2026 redemption surge - all four majors hit: Blue Owl OCIC 21.9%, OTIC 40.7%, Ares 11.6%, Apollo 11.2%, Blackstone 7.9%. Apollo returned just 45 cents per dollar requested. Moody’s cuts Blue Owl rating to ‘Negative’ Industry total: $20.8B (CNBC, CNBC, CNBC, Reuters, Reuters, Stanger, Wealth Management)\n\nMedian listed BDC is trading at just 0.73x its net asset value (NAV), the lowest since 2020 (Kobeissi Letter, March 2026)\n\nBlackRock TCP Capital: writedowns cut NAV by 19% in Q4 2025 - valuations catching up to reality (Wealth Management)\n\nTricolor indicted (Dec 2025) - DOJ alleges $800M double-pledged collateral. JPM $170M hit. First Brands: $2B+ unaccounted (CNBC)\n\n\"When you see one cockroach, there are probably more.\" - Jamie Dimon, JPMorgan CEO (Oct 2025) . \"The credit cycle has not been repealed.\" - David Solomon, Goldman Sachs CEO (Mar 2026)\n\nDetailed analysis: March 2026 - The $2.7 Trillion Shadow Lending Market Is Showing Cracks\n\nQ1 2026 data is in. Every major manager hit. The slow-motion bank run is now quantified.\n\nHormuz Reopened. Futures Fell 12%. The Damage Is Still Here.\n\nMarket celebrates, but 40 days of closure doesn't reverse in a day.\n\nIEA: even with full reopening, oil flows not expected to reach pre-conflict levels till late 2026 IEA\n\nKuwait Petroleum CEO: \"full production within three or four months.\" Capacity doesn't switch back on at a press conference CNBC\n\nTanker fleet mispositioned - pointed toward the US. Redirecting back to the Gulf takes months. Windward\n\nPump prices: \"months if not a full year\" to return to pre-war levels. Newsweek\n\n“This is a historic disruption to world oil. There has never been anything of this scale. Even the oil crises of the 1970s, the Iran-Iraq war of the 1980s, Iraq's invasion of Kuwait in 1990 - none come close to the magnitude of this disruption.” Daniel Yergin, vice chairman of S&P Global CNBC\n\n“The current oil and gas crisis triggered by the blockade of the Strait of Hormuz is more serious than the ones in 1973, 1979 and 2022 together.” Fatih Birol, IEA Head | Reuters\n\nAnalysis: \"The Oil Price You See Is Wrong\" (Apr 12, 2026)\n\nHow a Blocked Strait Hits Economies\n\nThe closure doesn’t need to last forever to cause lasting damage. It just needs to last long enough for these transmission chains to embed in prices. That process has already started.\n\nHormuz - Three Scenarios – One Thing in Common\n\nWhat Kills This Thesis (1/2)\n\nEach item has an observable trigger - track these, not the narrative.\n\nFed cuts aggressively + credit stress flattens. 75bps+ by Sept (not the 25bp drip priced in) and consumer 30+ DPD flow rates turn down from 4.8%. Hormuz relief is easing inflation - could give Fed cover. Apr 28-29 FOMC is the first test\n\nLabor market stabilizes without rolling over. 35-month drift (3.4% to 4.3%) unprecedented without recession. But JPM: ‘weakness from subdued hiring, not widespread layoffs’ Non-farm payrolls above 150K for 3+ months begins invalidating this leg\n\nPrivate Credit redemptions ease + BDC NAV recovers. Q1 requests ran 8-40% across majors; BDCs at 20% discount to NAV. If Q2 requests fall below 5% and price/NAV recovers toward 0.85x+, the bank-run dynamic breaks. No sign of easing yet\n\nAI productivity shows up in wages, not just layoffs. BofA data: K-shaped compression (upper +5.6%, lower +1%). Real wages positive across all brackets for 2+ quarters weakens consumer stress story. No data supports this yet\n\nConsumer savings rebuild above 5%. Student loan delinquency flow at 16.35% - highest since 2009. If oil normalizes, inflation falls, and savings rate rebuilds, the default transmission closes. Possible but needs 2-3 quarters of data\n\nWhat Kills This Thesis (2/2)\n\nThe Bull Case Has Data Behind It“Private credit probably does not present a systemic risk.” Jamie Dimon | Shareholder Letter, Apr 2026.\n\nCDS markets aren't alarmed. JPMorgan 5yr CDS at ~45bps today (52-week range 37-67bps), well below crisis-era peaks when IG CDX hit 280bps. Big-bank CET1 ratios now ~15-16% vs ~5-7% pre-2008 - roughly 2.5-3x thicker buffers. Credit pricing says \"stress, not crisis“\n\nPrivate Credit too small to be systemic. $2T vs $13T public bonds. Morgan Stanley: even 8% defaults \"significant but not systemic.\" IMF Apr: \"contained“\n\nFiscal stimulus flowing. One Big Beautiful Bill Act tax cuts live (signed July 2025, bulk active 2026). Hyperscalers committed $600B+ AI capex 2026 - nearly double 2025. Bessent (Apr 16): \"None of our work has shown there would be a systemic problem“\n\nRecession probability fading. J.P. Morgan: \"We no longer see a US recession.\" RSM US cut 40% to 30%. National Association for Business Economics GDP revised up. Same consensus that missed 2008\n\nThis deck isn't certainty - it's asymmetry. If three invalidators print next two quarters, the 2007 mirror shatters.\n\nGradualism Plus A New Variable\n\nCredit cycles have had slow burns before. 2006-07 is the reference case. The setup is same - delinquencies creeping, marks held at par, markets at highs.\n\nWhat has no precedent: a structural labor shift happening inside the same cycle.\n\nAI displacement is not cyclical. Jobs lost to AI don't return when credit stabilizes\n\nTwo pressures running simultaneously with no historical playbook\n\nThe slow pace gives time to react. Complacency is the risk. When it breaks, it breaks with two engines - credit and structural labor - not one\n\nThe economic cycle is familiar.\n\nThe AI displacement is not.\n\nData & Tools\n\nPrimary Data Sources\n\nNY Fed Consumer Credit Panel / Equifax\n\nFFIEC Call Reports via FDIC\n\nFDIC Statistics on Depository Institutions\n\nSNL Global Market Intelligence (auto ABS, subprime)\n\nBLS (unemployment, wages), SEC/EDGAR (BDC filings, 10-K/10-Q)\n\nLive Interactive Tool\n\nTREMOR - tremor.tigzig.com\n\nUS-Banks-Aggregates\n\nUS-NDFI Module\n\nMacro Charts\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: Market tools hub, Analysis decks, TREMOR indicators API\n\nMore posts: The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top., S&P at 7,400. Irrational Exuberance. The Last Two Times This Setup Held, the Index Halved., Index at a Record. Peak Valuation. Record-Low Sentiment. Last Two Times This Setup Held - 2000 and 2007 - the Market Halved., The Crisis in Auto Loans. 2007 was Sub-Prime Mortgage. Is 2026 Auto?, Three Red Flags Hit Together. 30Y at 5.14%, PPI at 5.99%, Brent Above $100. Setup Worse Than 2000 or 2007."
},
{
"slug": "sp500-drawdown-qrep-analysis",
"title": "S&P 500 Drawdown Analysis with QREP",
"tags": [
"portfolio-analytics",
"technical-analysis"
],
"content": "S&P 500 Drawdown Analysis with QREP\n\nS&P 500 is going through a drawdown. How many drawdowns in the past 20 years? How long? How deep?\n\nQREP pulls out drawdown charts and metrics for exactly this kind of analysis. One click.\n\nPowered by QuantStats. Free. Open source. Live at qrep.tigzig.com\n\nAlso get interactive technical analysis charts on the fly, plus performance, risk, trading KPIs and advanced ratios. Compare up to 6 securities in one shot.\n\nS&P 500 Drawdown Analysis\n\n\nRelated\n\nTools: QRep - Security Reports, Quants Agent, Quants Suite\n\nExplore: Quant Apps hub, MCP servers and agents, Analysis deck\n\nMore posts: QRep - Powered by QuantStats. Live Now., That 9X return from Nifty Midcap is irrelevant if you couldn't survive the 73% of time it was in drawdown, Gold up 2.3X past 2 years. But what about the 10 years drawdown in between., Three Tools I Built for Market Analysis - QSUITE, TREMOR, QREP"
},
{
"slug": "sp500-excel-tracker-xlwings-yfin-aug2026",
"title": "An Excel Tracker for Hundreds of Yahoo Finance Tickers, Built with xlwings Lite. For When WEBSERVICE Keeps Failing.",
"tags": [
"excel",
"xlwings-lite",
"python-in-excel"
],
"content": "An Excel Tracker for Hundreds of Yahoo Finance Tickers, Built with xlwings Lite. For When WEBSERVICE Keeps Failing.\n\nIf you are pulling hundreds of Yahoo Finance tickers into Excel through the Tigzig YFIN API and WEBSERVICE keeps failing, here is a tracker built for exactly that. Built with xlwings Lite, Python running inside Excel. Pulls dividend & earnings calendar data, and prices.\n\nUses YFIN's batching, up to 25 tickers at a time, batch size is configurable. Each batch writes back to the sheet as it comes in, with a spinner and a running status cell so you can see the load in progress.\n\nAll dates are UTC calendar dates in YYYY-MM-DD, no time of day. Yahoo timestamps each event to an exact instant, and we report the UTC date of that instant. For a US company reporting after roughly 8 PM Eastern, that instant crosses midnight UTC, so the UTC date is the next day. Convert from UTC to your own timezone rather than assuming the date is the US trading date.\n\nCurrently set up for S&P 500 tickers, swap in your own list. Full S&P 500 calendar pull takes around 9 minutes, price pull for 4 trading days takes around 2 minutes.\n\nxlwings Lite, built by Felix Zumstein, is free, comes as an Excel add-in you install once, and then you write and run Python inside Excel itself. It is my tool of choice for Excel automation. Installation is simple, guide linked below.\n\nIf you need to modify anything, the code is right there, hand it to any AI, including ChatGPT, and it should be able to change it for you. Guides and templates on xlwings Lite are on my hub, also linked below.\nLink to Tracker: SP500TRACKERPULL.xlsx\nInstallation and official docs: lite.xlwings.org/installation\nTigzig xlwings-Lite hub: xlwings-lite.tigzig.com\nTigzig YFIN API docs (humans): tigzig.com/apis/yahoo-finance\nTigzig YFIN API docs (AI): yfin-h.tigzig.com/v1/openapi.json\n\nDemystifying Tigzig API errors\n\nThe most common errors I see on my API backend and how to fix them: Demystifying Tigzig API errors\n\nAnd if you share your error along with the link to the API docs above with your AI, it should be able to identify and fix it for you. Full catalogue of all Tigzig APIs is available at:\nHumans: tigzig.com/apis\nAI: api.tigzig.com\n\nFor MFPRO API users\n\nFor MFPRO API (tigzig.com/mfpro) users reading this: the same setup works for the India mutual fund APIs too. Hand the base code here to your AI along with the API doc links above, and it should be able to set it up for you fairly quickly. In fact you can go one step ahead, and have it set up calculations for risk, return, benchmark and other KPIs too.\n\n\nRelated\n\nTools: DUCKIT - CSV to DuckDB, QRep - Security Reports, Quants Agent\n\nExplore: xlwings Lite hub, Yahoo Finance API docs, API + MCP hub\n\nMore posts: Getting 429s on the YFIN API from an Excel WEBSERVICE Formula? Here Is What Is Happening., Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users, TigZig API Rate Limits Raised Substantially, Across the Edge, the Proxy and the Apps. And a 429 Now Tells You What to Do About It., Claude Cowork Runs a Real Python Sandbox. I Put 37 Million Rows of Mutual Fund NAV Data Through It."
},
{
"slug": "sp500-irrational-exuberance-7400-may2026",
"title": "S&P at 7,400. Irrational Exuberance. The Last Two Times This Setup Held, the Index Halved.",
"tags": [
"portfolio-quants"
],
"content": "S&P at 7,400. Irrational Exuberance. The Last Two Times This Setup Held, the Index Halved.\n\nPublished: 2026-05-17\n\nS&P 500 at 7,400 with Irrational Exuberance. Last two times setup held: 2000 dot com -49% in 30 months, 2007 subprime -57% in 17 months. Apply pattern math, S&P lands 3,200-3,800. CAPE 39.6 (2nd time in 145 years), auto delinquencies past GFC peak, U-6 back to pre-GFC, PPI 25-year high ex-COVID, Brent $106. Dimon: too much exuberance. Buffett on $397B cash.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/sp500-irrational-exuberance-7400-may2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/sp500-irrational-exuberance-7400-may2026 with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/IRRATIONALEXUBERANCEMAY2026.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nTigzig Analysis Note - 17 May 2026\n\nExtraordinary Popular Delusions and the Madness of Crowds\n\nMarkets going up. Economy cracking underneath. S&P 500 @ 7,400 - 3,750? The brief documents what the bull case leaves out: valuations at a 155-year high, NVIDIA carrying the rally, credit cracking across auto, cards, and mortgages, PPI at a 25-year ex-COVID high, Brent at $109 with gas at $4.50, U-6 unemployment at pre-GFC levels, 3,642 Presidential trades filed past deadline, and warnings on record from Jamie Dimon and Warren Buffett.\n\n\"Men, it has been well said, think in herds; it will be seen that they go mad in herds, while they only recover their senses slowly, and one by one.\"\nCHARLES MACKAY - Extraordinary Popular Delusions and the Madness of Crowds, 1841\n\nWhere We Are\n\nTwo prior peaks. Each gave back roughly half. Today sits at the second-highest valuation in 145 years.\n\n[S&P 500 daily chart, Jan 1995 to 15 May 2026. Peaks marked at Mar 2000 (1,527), Oct 2007 (1,565), and today (7,408).]\n\nThe 2000 and 2007 peaks each preceded roughly a 50% drawdown in the index. Today's print of 7,408 sits at the second-highest CAPE valuation in 145 years. The rest of this brief lays out what stacks behind that number.\n\nSource: S&P 500 daily close (Jan 1995 to 15 May 2026), via Tigzig TREMOR Platform\n\nThe Bull Case\n\nThe bull case isn't crazy. It just isn't the whole picture.\n\nQ1 2026 S&P 500 earnings growth came in at 27.7% against an analyst consensus of 13.0% - more than double the estimate, and the highest blended growth rate since Q4 2021. Earnings, on the face of it, are strong.\n\n27.7% - actual\n13.0% - estimate\n\nThe case extends past earnings to two structural claims:\nThe US economy is far less oil-dependent than in past cycles - but gasoline at $4.50 a gallon, up 53% from pre-war levels, says otherwise.\nAI capital expenditure could top $1 trillion in 2027, roughly 3% of US GDP - but the cash flow profile is increasingly stretched, and the spend carries real redundancy risk from technological obsolescence.\n\nIf you stop at the headline, the index at 7,400 is just a richly valued market that earned its multiple. End of story. The rest of this brief is what the bull case leaves out.\n\nSource: CNBC, The stock market isn't ignoring Iran (12 May 2026); Fortune, Big Tech AI Capex Tracker 2026 (29 April 2026); CNBC, AI Boom: Big Tech Capex to Top $1 Trillion in 2027 (30 April 2026); FactSet Insight, S&P 500 Earnings Season Update (8 May 2026)\n\nIndex Concentration\n\n7 stocks. 35%+ of the index. The S&P 500 is a Mag 7 ETF in a trenchcoat.\n\nThe S&P 500 has always been market-cap weighted, but the top has not been this concentrated since 1932. According to Goldman Sachs, this level of single-name concentration has occurred only a handful of times in the past century - 1932, 1939, 1964, 1973, 2000, 2009, 2020, and now. Seven companies now make up more than a third of the entire index. The bottom 493 names share what is left.\n\n[SPY treemap: SPDR S&P 500 ETF (SPY) holdings, 14 May 2026. Mag 7 in red occupies the entire left half of the index by market cap. State Street Global Advisors.]\n\nThe picture is the point. The block of red on the left is seven companies. The mosaic of small blue tiles on the right is the other four hundred and ninety-three.\n\n~9% - NVIDIA's share of the S&P 500\n~35% - Mag 7 combined\n69% - of S&P 500 rally gains came from 10 stocks (late March - early May 2026)\n\nThe textbook case for buying the S&P 500 is diversification. You buy one fund and you own five hundred companies across every sector. Today, that is no longer mechanically true. Buy SPY and roughly a third of your money goes to seven AI-adjacent technology names.\n\nThe index is not the economy. It is a concentrated handful of AI names.\n\nSource: SPDR S&P 500 ETF (SPY) holdings, State Street Global Advisors (14 May 2026); Investing.com, The S&P 500's Narrow Rally: 10 Stocks Accounted for 69% of Gains; Goldman Sachs, Is the S&P 500 too concentrated?\n\nThe Trump Trades\n\n3,642 trades in Q1. Several preceded regulatory or trade actions for the same companies.\n\nThe Office of Government Ethics filing is public record. Form 278-T, the periodic transaction report covering Q1 2026, lists 3,642 individual transactions in the President's name, filed 8 May, past the OGE's statutory deadline of 45 days. The January trades sat undisclosed for 90+ days before filing.\n\nThe pattern that catches the eye is the proximity of certain purchases to specific federal actions on the same companies.\n\nJAN 6 - NVIDIA purchased\n1 WEEK LATER - Commerce approves resumption of NVIDIA chip sales to China.\n\nJAN 6 - AMD purchased\n1 WEEK LATER - Commerce approves resumption of AMD chip sales to China.\n\nFEB 10 - AXON purchased\n2 WEEKS LATER - ICE announces $220M Taser contract awarded to Axon.\n\nJAN - MAR - Boeing purchased (multiple)\nMAY 14, 2026 - On his Beijing trip, Trump tells press of a major Chinese aircraft order for Boeing. No signed agreement has been disclosed.\n\nEach individual transaction may have an innocent explanation. A diversified portfolio held in trust will buy and sell large names routinely. Regulatory and trade actions move through the executive branch on their own schedule. Coincidence in markets is more common than people think.\n\nWhether any individual trade is \"informed\" is impossible to determine from the outside. What is harder to chalk up to coincidence is a pattern, repeated across multiple positions, over months, originating from the office that shapes the calendar of the actions that followed.\n\nThe case that the broader market should be uncomfortable is not a verdict on any one trade. A congressional or SEC probe puts these names under direct overhang. And the Mag 7 premium that prices in a friendly administration starts looking borrowed.\n\nThe premium that prices in a friendly administration is only as good as the administration's appetite for the inquiry.\n\nSource: OGE Form 278-T, Donald J. Trump Q1 2026.pdf) (filed 8 May 2026); NOTUS (May 2026); CNBC, Trump went big on tech stocks in first quarter of 2026, new filings show (15 May 2026); CNBC TV18, Donald Trump Trading and Investment Disclosures Q1 (May 2026); CNBC, Trump wraps up two-day China trip (15 May 2026)\n\nValuation at 39x\n\n39x has only happened twice in 145 years. Dot-com peak. Now.\n\n[Shiller CAPE ratio chart, January 1881 to May 2026. Today's reading of 39.6 sits second only to the December 1999 dot-com peak of 44.2.]\n\nThe Shiller CAPE ratio, cyclically adjusted price-to-earnings, is the long-term valuation measure Robert Shiller has tracked since 1881. It sits at 39.6. The long-term mean over the full 1881-2026 series is 17.7. We are trading at a 123% premium to that mean.\n\nTo put the number in context: in October 2007, the month before the S&P 500 began its 57% decline into the Global Financial Crisis, CAPE was 27.5. Today's reading is 44% higher than that pre-GFC peak.\n\nWhere today's reading sits\n\nDOT-COM PEAK - Dec 1999 - 44.2\nTODAY - May 2026 - 39.6\nPRE-GFC PEAK - Oct 2007 - 27.5\nLONG-TERM MEAN - 1881-2026 - 17.7\n\nThere has been exactly one period in the entire 145-year series where the market traded higher than today on this measure. It was December 1999 to August 2000. The Dot-com peak. The crash that followed cut the S&P 500 by 49% over 30 months.\n\nA high CAPE doesn't mechanically forecast a crash. It does say one thing reliably: forward returns from this starting point have, historically, been poor. That is what the long series tells us.\n\nThe data won't tell us when the air leaves. It tells us how much air is in.\n\nSource: Robert Shiller, Online Data, Yale University (CAPE Ratio) via tremor.tigzig.com\n\nOil Shock\n\nBrent at $106. Gas at $4.50, up 50% since February. Ceasefire is priced in. Supply isn't.\n\nBrent crude futures sit at $109 on 15 May. Pre-war they were at ~$70. Spot prices touched $138 during the active conflict, with futures peaking near $114 in early May (chart below), and the post-ceasefire pullback has been shallow. Brent now trades roughly 50% above its pre-war baseline. US gasoline at the pump averages $4.50 a gallon, up 53% from the $2.94 pre-war baseline. California is past $6.\n\n[Brent crude oil futures chart, daily, May 2016 to May 2026. Current $109 (15 May). Pre-war level was ~$70. Futures peaked near $114 in early May (spot touched $138).]\n\nThe market has chosen to believe the ceasefire announcement closes the file. Equities have priced in a return to status quo. The physical supply chain has not.\n\nTransit time alone is four to six weeks. Tankers move at 12 to 15 knots and depots are not next door to refineries. The longer tail is infrastructure: pipelines, terminals, and processing capacity damaged during the conflict do not rebuild in a month. The IEA's read is that flows do not return to pre-conflict levels until late 2026. So the gap between \"war is over\" and \"barrels are landing at scale\" is weeks for transit, months for restoration, stacked on top of each other.\n\nUS Gasoline Prices - Three Crosses of $4 in 25 Years\n\n[Monthly US national average gasoline price (regular unleaded). Source: FRED via Tigzig TREMOR Platform.]\n\nNo US price is watched more closely by American politics than gasoline at the pump. Historically, when gas crosses $4 a gallon, household sentiment drops and consumer spending on goods softens within one to two quarters. The pass-through that PPI is already warning about gets a second leg.\n\nThe 2008 cross of $4 came in June and lasted about two months. University of Michigan consumer sentiment in May 2008 hit its lowest level since the 1980 recession. The 2022 cross in March lasted about six months, and real consumer spending on goods weakened through the second quarter. In 2022 the policy response was substantial: the largest SPR release in history, plus state-level gas-tax holidays. In 2008 it was mostly rhetoric, with a federal gas-tax holiday debated but never enacted. The pattern in both episodes is the same: by the time policy catches up, the damage to consumer behaviour has already shown up in the data.\n\nThe bull-case framing of a contained inflation environment requires this to reverse fast. Tankers do not move fast.\n\nThe ceasefire is priced into the equity market. The barrels are not yet on the water.\n\nSource: FRED US Average Gasoline Price (Monthly Historical) via Tigzig TREMOR Platform; CNBC, Gas Prices Hit National Average of $4 a Gallon (8 June 2008); FRED, University of Michigan Consumer Sentiment (Historical Series); AAA, National Average Exceeds $4/Gallon (2 April 2026); BEA, US Consumer Spending - Quarterly Real PCE\n\nOil Scenarios\n\nWar started Feb 28. Hormuz still effectively closed. 75+ days and counting.\n\nThree forward paths. None of them clean.\n\nScenario 1: Ceasefire holds, Hormuz reopens\n3-5 months to restore supply. Brent by Q3: $80-90. Painful but manageable.\n\nScenario 2: Ceasefire holds, Hormuz restricted\nShips trickle through with approval. Brent through 2026: $100-120. Recession risk rises materially.\n\nScenario 3: Ceasefire collapses\nFull blockade. SPR mostly spent. Brent futures: $130+. Stagflation base case.\n\nThe next 2-6 months are tight in all three scenarios.\n\nTankers move at 12-15 knots. Reservoirs take weeks to restart. Destroyed infrastructure will take years to rebuild.\n\n75+ days. Hormuz still effectively closed.\n\n\"We are facing the biggest energy security threat in history.\"\nFatih Birol, IEA Executive Director - April 2026\n\nSource: CNBC, Oil Markets & Iran War, IEA Chief Fatih Birol (23 April 2026)\n\nThe Inflation Warning\n\nPPI at 5.99% YoY. Highest since the 2022 peak. Margins or prices, somebody pays.\n\nThe April 2026 Producer Price Index, Final Demand, came in at 5.99% YoY. That is the highest reading in 25 years with one exception: the 2022 post-COVID supply-chain spike that peaked around 11%. Outside that one episode, today's number is the hottest the PPI has run since the early 2000s.\n\nThe PPI is what producers pay. It does not stay at the producer level. It either gets absorbed (margins compress) or it gets passed forward (consumer prices rise). Both outcomes have macroeconomic costs. The cycle from PPI to CPI typically runs three to nine months.\n\n[US PPI Final Demand chart, year-over-year (%). The 2022 peak of 11% was the COVID-era one-off. The 5.99% April 2026 print is the highest reading outside that spike in the entire 25-year series.]\n\nThe bull-case framing for equities prices in stable margins and contained inflation. Neither is consistent with PPI at 6%. Companies will defend earnings by passing costs through. Consumers, whose credit data is already cracking, are not in a position to absorb another wave.\n\nProducer pain does not disappear. Either margins compress, or it flows to the consumer. There is no third option.\n\nSource: US Bureau of Labor Statistics, Producer Price Index Final Demand (April 2026), via Tigzig TREMOR Platform\n\nThe Labor Crack\n\nU-6 on a steady rise since late 2023. Now at pre-GFC 2007 levels.\n\nMost market commentary refers to \"the unemployment rate\" without specifying which one. The headline rate, U-3, counts people actively looking for work and not finding it. It is currently at 4.3%. The broader measure, U-6, adds two more groups: people working part-time because no full-time job is available, and people who want work but have stopped actively looking. U-6 is the rate that captures stress that has not yet shown up in the official tally.\n\nU-6 sits at 8.2% as of April 2026. The headline U-3 at 4.3% and the broader U-6 at 8.2% are both within touching distance of where they stood in October 2007, one month before the S&P 500 began its 57% decline into the Global Financial Crisis.\n\n[US U-6 underemployment rate chart, monthly. The 2007 reading was ~8.4% just before the GFC. The 2026 reading is 8.2%, recovered from the post-COVID low and climbing steadily since late 2023.]\n\nThe trajectory matters more than the level. U-6 troughed at 7.2% in late 2023 and has climbed steadily since. That is two and a half years of slow upward drift, the kind of pattern that historically signals labor-market slack accumulating beneath an unchanged headline. The next section will show consumer credit doing the same thing on a different clock - cracking quietly while official statistics tell a calmer story.\n\nCompare the shape to 2006-2007. U-6 troughed in 2006, drifted higher through 2007, then snapped vertical in 2008. The pattern of patient accumulation, dismissed by markets at the time, was the early read of a recession already underway. The current chart looks like the early phase of that same shape.\n\nHeadline unemployment is the last data series to confirm what the broader measures have been signalling for two years.\n\nSource: US Bureau of Labor Statistics, U-6 Total Unemployed & Marginally Attached (April 2026), via Tigzig TREMOR Platform\n\nThe Consumer Credit Crack\n\nAuto past GFC peak. Cards near it. Mortgage flow cracking. All before the recession.\n\nThe standard story is that consumer credit stress shows up after a recession. The household balance sheet absorbs shocks, then breaks. By the time delinquencies appear at scale, the labor market has already turned. This cycle has flipped the order. Three segments are flashing pre-recession patterns simultaneously, and one has already crossed its 2007 peak.\n\nA reading note. The NY Fed 90+ DPD rate used here is broader than the bank-portfolio definition - it includes accumulated severely derogatory balances (charge-offs, repossessions, foreclosures kept on the credit file for up to 7 years). Closer to a credit-loss measure than a current-delinquency measure. Read \"3.4% past the 3.1% pre-GFC level\" as a loss-rate proxy crossing a historically dangerous threshold.\n\n[Auto loans 90+ DPD chart: 5.6%, past the 5.3% GFC peak. The first major consumer segment to break beyond GFC levels.]\n\nThe mortgage flow signal\n\nThe chart that matters here is the one that looks like the early subprime tape. The stock rate, total 90+ DPD as a share of all mortgage balance, still looks tame near 1%. CARES Act forbearance distorted that trough and the climb off pre-pandemic levels has been slow.\n\nThe signal that matters is the straight-through flow rate. It captures the share of balance that was current at the end of last quarter and is now 90+ days past due this quarter. To get there, the borrower had to miss three consecutive monthly payments with no recovery in between. Not a single missed payment that gets caught up. Sustained distress, all in one quarter.\n\n[Mortgage straight-through flow rate chart: 0.26%, approaching the 0.35% pre-GFC level. Sharp climb in the last six months. The early-crack signal that stock numbers miss.]\n\nThe pattern is reminiscent of the early subprime tape from 2007. Everything looked fine for years. Then the chart turned, abruptly, and kept going.\n\n[Credit cards 90+ DPD chart: 13.1%, closing on the 13.7% GFC peak. The climb to here happened without a recession. In 2008-09, it took one.]\n\nAcross the credit complex\n\nAcross all consumer debt, the 90+ DPD rate sits at 3.4%. Already past the 3.1% pre-GFC reading from Q3 2007.\n\n[All consumer debt 90+ DPD chart: 3.4%, past the Q3 2007 pre-GFC reading of 3.1%. The 2010 peak of 8.7% was reached only with a recession. This time, the line was crossed without one.]\n\nThe labor data is now confirming what the credit data flagged earlier. As shown in the previous section, headline U-3 and the broader U-6 are both at pre-GFC levels. The credit signal turned first; the labor signal is the second leg.\n\nA wave across consumer loans without a recession is one thing. Mortgage flow cracking on top of it is another.\n\nSource: NY Fed Consumer Credit Panel / Equifax, Q1-2026 update, via Tigzig TREMOR Platform; Tigzig, US Lending Crisis II (14 May 2026)\n\nThe Dimon Warning\n\nThe CEO who runs the largest US bank is on record. Last week.\n\nMarkets pay attention when the head of JPMorgan Chase, the largest bank in the United States by assets, is on a public panel.\n\n\"...a little bit too much exuberance out there. The stock market is in the top 15%, credit spreads are very low. The general assumption is that these things are all going to resolve. And I'm kind of a skeptic.\"\nJamie Dimon, CEO JPMorgan Chase, May 2026\n\nSource: Bloomberg Television, JPMorgan's Dimon Sees Too Much Exuberance in Markets (12 May 2026, interview video); TheStreet, JPMorgan doubles down on stock market message for 2026 (May 2026)\n\nThe Buffett Warning\n\nThe man with $397B in cash sees a casino, not a market.\n\n$397.4B - Berkshire cash pile, Q1 2026 - highest in 60 years\n\n\"We've never had people in a more gambling mood than now.\"\nWarren Buffett, Berkshire Hathaway 2026 Annual Meeting, May 2026\n\nSource: CNBC, Berkshire Hathaway 2026 Annual Meeting Transcript (2 May 2026); CNBC, Berkshire Hathaway 2026 (2 May 2026)\n\nHistory Repeating?\n\nMarkets at highs. Warnings dismissed. Cracks visible to anyone looking.\n\nThree patterns, three different decades. The four questions that defined each cycle are the same.\n\n2000 - Dot-com\nValuation: Shiller CAPE 44.2\nNarrative: .com / tech\nDismissed: No path to profit\nTrigger: Tech earnings missed\n\n2007 - Pre-Crisis\nValuation: Shiller CAPE 27.5\nNarrative: Subprime mortgage\nDismissed: Subprime cracking quietly\nTrigger: Subprime defaults\n\n2026 - Now\nValuation: Shiller CAPE 39.6\nNarrative: AI / Mag 7\nDismissed: Cracks across credit, labor, inflation\nTrigger: TBD\n\nThe point is not that 2026 will mirror 2000 or 2007. The point is that the pattern is the same: a richly-priced index, a hot narrative carrying the rally, structural cracks dismissed because they have not yet broken, and a consensus that today is different.\n\nEach prior cycle had its own surface story. By the time the surface story changed, the index had given back half its value or more.\n\nEach prior cycle thought it was different. Each prior cycle was right - the surface story was different. The underlying pattern was not.\n\nSource: Shiller CAPE Data, Yale University; NY Fed Q4 2025 Household Debt and Credit Report (10 February 2026); Historical S&P 500; FCIC archives\n\nThe Full Bear\n\nThe last two times we were here, the index halved.\n\nThe previous section laid out the structural parallel: 2000, 2007, 2026, three cycles with the same pattern. This section takes that pattern and runs the arithmetic forward. From peak to trough, in both prior cycles, the S&P 500 lost roughly half its value. The exact numbers:\n\n2000 - Dot-com\nPeak: 1,527 - Trough: 777 - -49% over 30 months\n\n2007 - Pre-Crisis\nPeak: 1,565 - Trough: 677 - -57% over 17 months\n\n2026 - If repeats\nPeak: 7,400+ - Trough: ~3,700 - ~-50% over TBD\n\nThe 2000 cycle took thirty months to find its bottom. The 2007 cycle took seventeen. Each ended at roughly half of where it began. Apply the same math to a 7,400 print today and the implied trough sits around 3,700.\n\nThis is not a prediction. It is what the math of the last two cycles gives, applied to today's starting point. The cycles do not have to follow the same script. Sometimes the index loses less; sometimes it loses more.\n\nNot a forecast. The math of the last two cycles, applied to today.\n\nSource: Historical S&P 500 monthly close data; NBER recession dating; Shiller CAPE Data, Yale University\n\nThis is historical pattern analysis. Not a price target or forecast.\n\nWhat Proves This Wrong?\n\nNone of this is certainty. It's asymmetry. Here's what would prove it wrong.\n\nA position taken without a kill criterion is a story, not a thesis. The pattern in this brief, valuation at the second-highest in 145 years, credit cracking under an unchanged headline, oil and inflation re-stacked on top of a stretched consumer, names with overhang from outside the market, is an asymmetric bet against the consensus. It is not a forecast. Markets can absorb worse setups than this and still go higher.\n\nThe cleanest way to hold a view like this is to name, in advance, the data points that would force a reversal. The five below are mine.\n\n1 - OIL\nHormuz reopens fully and Brent drops below $80 within Q3 2026. The inflation section's headline number rolls over with it.\n\n2 - EARNINGS BREADTH\nMag 7 ex-NVIDIA earnings re-accelerate above the Other 493 by Q2 reporting. The concentration story stops being a one-name story.\n\n3 - CONSUMER CREDIT\nDelinquency flow rates turn down across auto, cards and mortgage simultaneously. Not one segment improving while another deteriorates - the entire complex rolling over.\n\n4 - PRIVATE CREDIT\nNo contagion from the next disclosed loss event. A material write-down at a major BDC absorbed without forced selling, gating, or redemption-driven mark-to-market spirals.\n\n5 - AI CAPEX RETURNS\nThe trillion-dollar capex cycle translates to broad productivity gains and wage growth in non-AI sectors, not concentrated layoffs in everything adjacent.\n\nThese are not predictions in reverse. They are tripwires. Each one, on its own, would weaken the thesis. Two or three together would invert it. None of them alone, however, can be ignored when they print.\n\nFive tripwires. None tripped yet.\n\nData & Tools\n\nThe numbers and charts in this brief come from primary US data sources, aggregated and tracked through the Tremor platform.\n\nPrimary Data Sources\nRobert Shiller, Yale University. Online Data, CAPE Ratio (1881-present).\nUS Bureau of Labor Statistics. Producer Price Index, Final Demand; U-6 Underemployment Rate.\nFederal Reserve Bank of New York. Consumer Credit Panel / Equifax. Quarterly Report on Household Debt and Credit.\nEIA / AAA. Short-Term Energy Outlook, Brent Crude prices, National Gas Average.\nUS Office of Government Ethics. Form 278-T disclosure filings.\nGoldman Sachs. S&P 500 concentration analysis.\nReporting: CNBC, Bloomberg, Fortune, TheStreet, FactCheck.org, CNBC TV18, and others.\n\nTremor Platform\n\ntremor.tigzig.com\n\nA macro early-warning dashboard tracking 100+ global macro indicators and 30+ news sources, plus US lending aggregates from FDIC, NCUA and the NY Fed. Every series carries three-layer validation with automated cross-checks. Per-chart customisation is available.\n\nMost of the charts reproduced in this brief - Shiller CAPE, Brent crude, gasoline, PPI, U-6, consumer 90+ DPD - are live views from Tremor.\n\n[Tremor Macro Stress Signals dashboard. Live charts across global indices, credit, and commodities.]\n\nPrevious Analysis\n\nThis is the latest in a series of analyses tracking US credit, banking, and market stress. Earlier notes:\n\nMAY 15, 2026\nUS Lending Crisis II - NY Fed Q1 2026\nAuto past GFC peak. Cards near it. Mortgage cracking at the flow. Without a recession.\n\nMAY 8, 2026\nRed Flag in US Life Insurance\n$2.4T (7.5% of GDP) in private credit and offshore reinsurance. Doom loop risk.\n\nAPR 29, 2026\nThe Convergence Risk\nCredit stress past 2007 pre-crisis. Five aggravators stacking on top.\n\nAPR 25, 2026\nUS Credit Unions on TREMOR\nCard charge-offs at 5.11% - already past the 2009 GFC peak.\n\nAPR 22, 2026\nThe Crisis in Auto Loans\nNY Fed Auto 90+ DPD at 5.21% vs 5.27% at GFC peak. Tricolor and First Brands fraud.\n\nAPR 18, 2026\nThe Disbelief Rally of 2026\nS&P at all-time high while private credit, consumer delinquencies and bank charge-offs above 2007 levels.\n\nAPR 12, 2026\nThe Masking Effect in Oil Markets\nLargest spot-futures divergence in oil market history. Hormuz shut 40 days.\n\nMAR 24, 2026\nPrivate Credit - The $2.7T Shadow Lending Market\nCracks reminding Wall Street of 2008. Dimon and Blankfein flagging risks.\n\nAll available at tigzig.com/analysis\n\nAbout the Author\n\nAmar Harolikar, ACA\n\nDecision Sciences & Applied AI - ex-Derivative Trader\n\nBuilder of tigzig.com.\n\nContact\namar@harolikar.com\n+91-98452-30763\nlinkedin.com/in/amarharolikar\n\nTools\ntigzig.com - tremor.tigzig.com\n40+ tools for analytics, quants, databases, dashboards & automation.\n\nAnalysis notes. Tools. Analytics. Published as I work.\n\nDISCLAIMER\n\nThis note is published for informational and educational purposes only. It reflects the author's analysis of publicly available data and does not constitute investment advice, a recommendation, or a solicitation to buy or sell any security. While reasonable care has been taken to ensure the accuracy of the information presented, the author makes no representation or warranty as to its completeness or correctness. The author shall not be liable for any errors, omissions, or losses arising from the use of this information. Readers should verify all data independently and consult their own financial advisors before making investment decisions. The author may hold positions in securities discussed.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: Market tools hub, Analysis deck series, Data APIs and MCP\n\nMore posts: Index at a Record. Peak Valuation. Record-Low Sentiment. Last Two Times This Setup Held - 2000 and 2007 - the Market Halved., S&P at All-Time High. Under the Hood, Things Are Cracking. Just Like October 2007., The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top., Three Red Flags Hit Together. 30Y at 5.14%, PPI at 5.99%, Brent Above $100. Setup Worse Than 2000 or 2007., BIS, the Central Bank of Central Banks, Just Red-Flagged the AI Boom. Railway Mania, Roaring 20s, Dotcom - Same Chart."
},
{
"slug": "sp500-vs-nifty50-returns-profile-reversing",
"title": "U.S. Markets (S&P 500) vs India (Nifty 50) - is the returns profile reversing?",
"tags": [
"portfolio-analytics"
],
"content": "U.S. Markets (S&P 500) vs India (Nifty 50) - is the returns profile reversing?\n\nS&P 500 vs Nifty 50\n\nNifty 50 has generally outperformed S&P 500 for a long time - going back to post-Y2K if my recollection is right. ...but there have been windows where the trend reversed.\n\nIs this one of them?\n\nI ran the numbers across multiple time horizons - starting 2007 before the 2008 crash (18 years) ....10 years, 5 years, 3 years, 2 years, and 1 year. The charts tell an interesting story. Long term, Nifty still leads. But zoom into the last 2-3 years and S&P has pulled ahead significantly.\n\nTemporary? Structural? You'll get as many opinions as there are experts. I'm still processing the data myself.\n\nRun your own reports at quants.tigzig.com - ask the agent ... ask for Security Performance Report for your securities and date ranges ... or hit 'Go to Suite' if you prefer a form (it's faster).\n\nDisclaimer: Educational analysis only, not investment advice.\n\n\nRelated\n\nTools: QRep - Security Reports, Quants Agent, TREMOR - Macro Stress Signals\n\nExplore: S&P 500 vs Nifty 50 - full slide deck (PDF), Quant Apps hub, Analysis hub\n\nMore posts: Nifty and S&P 500 Down - Run a Technical Analysis Check. 9 LLMs Compared., S&P 500 Drawdown Analysis with QREP, NIFTY50 - 30 Day Forward Return Analysis Feb 2008 to 2026 - Claude in Excel with Python, Lambdas and Advanced Formulas, Security Performance Report for Investors. AI Quant Agent. Live. Open Source. Free."
},
{
"slug": "spot-the-odd-chart-out-may2026",
"title": "Which Chart Is the Odd One Out? Three Macro Charts Tell the Same Story. One Doesn't.",
"tags": [
"portfolio-quants"
],
"content": "Which Chart Is the Odd One Out? Three Macro Charts Tell the Same Story. One Doesn't.\n\nPublished: 2026-05-31\n\nFour-chart US macro one-pager (May 2026) asking the spot-the-odd-one-out question. Three charts show rising stress - U-6 underemployment at 8.2% (climbing since 2023, up from 6.7% Jan 2023 and past 8.7% Peak 2025), personal saving rate down to 2.6% (from 6.4% peak 2024 and 4.9% Jan 2023, the cushion is nearly gone), consumer 90 DPD loss rate at 3.4% (past the pre-GFC 3.06% reference, worst at credit unions and non-bank lenders). The fourth chart: S&P 500 at a record 7,580, up 86% over the three-year window. Three macro-stress charts in one direction; the index chart in the other. Tools that drive this kind of comparison are on tigzig.com - free, no login, 300+ series covering macros, prices, savings, labor and delinquencies, available via web interface, API or MCP for AI agents.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/spot-the-odd-chart-out-may2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/spot-the-odd-chart-out-may2026 with header Accept: text/markdown\n\nTags\nportfolio-quants\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: Quant Apps hub, Analysis briefs, APIs and MCP\n\nMore posts: S&P at 7,400. Irrational Exuberance. The Last Two Times This Setup Held, the Index Halved., S&P at All-Time High. Under the Hood, Things Are Cracking. Just Like October 2007., The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top."
},
{
"slug": "stagflation-lite-macro-signals-march-2026",
"title": "Are We Headed for Stagflation-Lite",
"tags": [
"portfolio-quants"
],
"content": "Are We Headed for Stagflation-Lite\n\nPublished: 2026-03-28\n\nStagflation-Lite signals intensifying. Fed trapped between inflation and weak jobs. US 10Y at 4.4%, PPI at 3.4% with oil shock not yet in data, GDP at 0.7%, consumer sentiment 7th lowest in 40 years, Moody recession odds at 48.6%. Bond markets pricing in high inflation globally.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/stagflation-lite-macro-signals-march-2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/stagflation-lite-macro-signals-march-2026 with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/TREMOR28MARBONDS.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nAre we headed for Stagflation-Lite?\n\n1\n\nBond yields have spiked\n\n2\n\nConsumer sentiment near 40 year lows\n\n3\n\nUS Q4’25 GDP growth rate down to 0.7%\n\n4\n\nInflation expectation rising\n\n5\n\nOil, employment & delinquencies in a red zone\n\n6\n\nEquity markets in a micro-bear\n\nTREMOR –monitor early warning signals\n\nBuilt by analyst – for analysts\n\nCockpit view of key macros + News\n\nLive at: tremor.tigzig.com\n\nHere’s what the TREMOR charts are telling us\n\nBond markets pricing in high inflation expectation\n\nUS Past 20 Years\n\nPast 6 Months\n\nUS Treasuries 10Y Yield\n\nGermany 10Y Bund Yield\n\nInstitutional and bond desk expecting higher inflation. Which make the current yields unattractive, causing sell-off and pushing yields higher. Becomes self fulfilling – the expectation itself create outcome in bond markets increasing risk to real economy CNBC – Rising European Yields, Bloomberg Podcast\n\nUS Feb’26 PPI annual inflation at 3.4%, higher than expected\n\nUS Past 20 Years\n\nThe producer price index is a measure of pipeline costs that producers receive for their products and an early indicator of consumer inflation. The annual inflation rate was at 3.4%, highest in past year. And the oil shock is yet to be baked in\n\nConsumer sentiments running abysmally low\n\nMichigan Consumer Sentiment results released March 27 came in at 53.3 amongst the lowest reading (7th lowest) in past 40 years\n\nUS GDP growth rate revised to 0.7%\n\nThe first revision of GDP on13th March 2026 was a 50% lower than the previous 1.4% and below Dow Jones consensus forecast of 1.5% CNBC (Seasonally & inflation adjusted Annualized QoQ growth rate)\n\nOil, unemployment and delinquencies in red zone\n\nBrent continues to be above $100 and even when the Iran war ends and prices fall, the transmission won’t be instant. Some capacities may get online in months, others might take years. All while pressures continue on employment side and rising delinquencies.\n\nEconomists raising odds of a recession\n\nMoody’s Analytics’ model has raised its recession outlook for the next 12 months to 48.6%. Goldman Sachs boosted its estimate to 30%. Wilmington Trust has the odds at 45%, while EY Parthenon has it at 40%, with the caveat that “those odds could rapidly rise in the event of a more prolonged or severe Middle East conflict.”\n\nIn normal times, the risk for a recession in any given 12-month span is around 20%. So while the current predictions are hardly certainties, they signify elevated risk.\n\n“I’m concerned recession risks are uncomfortably high and on the rise,” said Mark Zandi, chief economist at Moody’s Analytics. “Recession is a real threat here.”\n\nRecession odds climb on Wall Street as economy shows cracks beneath the surface CNBC, March 25, 2026\n\nEquity markets entering micro-bear territory\n\nKey equities dropped ~10% the past month from their recent highs\n\nAre we heading for a stagflation like scenario?\n\nStagflation = High Inflation + Stagnant Growth\n\nExpectation of higher inflation\nExpectation of lower growth\nContinuing negative shocks\n\n=\n\nHigher probability of stagflation like scenario\n\nor 'Stagflation-Lite'\n\nPrevious Analysis\n\nAre we staring into face of a full bear market?US Banks & Non-Bank Lending - How Deep is the Exposure? Is there a Systemic Risk?Structural Weakness IntensifyingPrivate Credit - The $2.7 Trillion Shadow Lending Market Is Showing Cracks\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: Market tools hub, Analysis deep-dives, APIs and MCP\n\nMore posts: Tremor - Macro Early Warning Signals - March 2026, Three Red Flags Hit Together. 30Y at 5.14%, PPI at 5.99%, Brent Above $100. Setup Worse Than 2000 or 2007., There Is No Modern Playbook for This. US Core Inflation at 3.3%, Re-Accelerating. Producer Prices at 6.4%. Consumer Delinquencies Past 2007., The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top., TREMOR - Macro Early Warning Dashboard - 50+ Indicators, Updated Daily"
},
{
"slug": "stock-data-to-ai-reports-python-in-excel-xlwings-lite-part-4",
"title": "Stock Data to AI Reports | Python-in-Excel | xlwings Lite - Part 4",
"tags": [
"xlwings-lite",
"python-in-excel",
"technical-analysis"
],
"content": "Stock Data to AI Reports | Python-in-Excel | xlwings Lite - Part 4\n\nPull stock data, Run AI analysis, Get PDF reports - right from Excel.\n\nYahoo Finance Analyzer\n\nCheck out detailed info on on REX AI Co Analyst site\nFull templates\nHow-to guides\nBackend code\nTechnical breakdowns\nResource Links\n\nDetailed info, source codes, how-to guides, resources and templates available on REX\n\nxlwings Lite: Pushing the boundaries of Excel\n\nReleased just last week by Felix Zumstein (creator of the original xlwings). Based on what I've tried hands-on so far, it completely redefines what is possible in Excel, and I don't say that lightly. A mini code editor, console logs, Web API access - directly inside Excel. A single add-in install. And free. Processing, automation, databases, fin/stats analysis, machine learning - handles that out of the box. Tight integration with Excel object model. Going beyond that - build full tools inside Excel tied into APIs, databases, LLMs, n8n, Flowise, PDF generators, and agents. And I've barely touched the surface.\n\nIn this example\n\n… with Python scripts, running inside Excel…\nPull data from Yahoo Finance\nCompute technical indicators\nSend to Gemini for insights\nGet back a PDF-formatted technical analysis report\nPull financials - P&L, B/S, Cash Flows, Qtrly - multi-period\n\nall in seconds…\n\nTry it out\n\nInstall xlwings Lite Add-in (Excel > Home > Add-in > search \"xlwings lite\").\nGet a Gemini API key. Needed only for the AI report. Free, takes less than a minute. Go to aistudio.google.com → Get API Key → Create AI Key → Select default project.\n\nTemplates to explore\nMonte Carlo simulation with dashboards - great share by FRANCO SIVILA: https://lnkd.in/ghAsXgW5\nWeb API calls, DB connect - explore, query, pull data, analyze https://lnkd.in/g7mEJfVT\nAPI calls to LLMs - schema detection, EDA, ML https://lnkd.in/gXF4PEvV\n\nCode and build guides\n\nGitHub repo & Full video (8 mins) with process and code walkthrough: Go to Docs\n\nFine Print\n\nYLENS: Yahoo Finance Analyzer is not investment research or advice. It's a working prototype to demonstrate what's possible with xlwings Lite. AI and automation setups in live use cases require iteration, validation, and human judgment.\n\nExplore 15+ open-source AI analytics and automation apps at tigzig.com\nMutual Fund Disclosures Excel to CSV -AI Automation with Schema detect -https://lnkd.in/gxVbCdpv\nConnect ChatGPT to n8n for automation - tracker updates, PDF reports, slides & Google Apps Script automation https://lnkd.in/gTDC74gY\nAdvanced Analytics Multi Agents- https://lnkd.in/g8xPQvb8\n\nVideo Guide\n\nDemo, architecture and code walkthrough\n\nhttps://youtu.be/nnsO8XmLYuk\n\nExcel template, FastAPI server Github Repos and how-to guides: Download here\n\nxlwings Lite official site: lite.xlwings.org\n\n\nRelated\n\nTools: Stock Data to AI Reports, Technical Analysis in Excel, Practice Lab\n\nExplore: AI in Excel hub, Yahoo Finance API doc, Quant Apps hub\n\nMore posts: AI + Python in Excel with xlwings Lite - LLM API Calls | Part 3, xlwings lite |Connect to Remote Databases, AI-Powered Technical Analysis in Excel - with Python & Gemini Vision"
},
{
"slug": "structural-weakness-intensifying-march-2026",
"title": "Structural Weakness Intensifying",
"tags": [
"portfolio-quants"
],
"content": "Structural Weakness Intensifying\n\nPublished: 2026-03-20\n\n3 major US macro data points in 3 days all pointing the same direction. Feb payrolls minus 92K against expectations of 59K rise. PPI Core at 3.87%. Fed held at 3.5-3.75%. Brent crude nearly doubled in 3 months. Delinquencies past COVID highs. Multiple macro signals flashing stress simultaneously.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/structural-weakness-intensifying-march-2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/structural-weakness-intensifying-march-2026 with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/TREMORMAR2026_20MAR.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nYou don’t know who’s been swimming naked until the tide goes out. - Buffet\n\nGhosts of 2008\n\nIs Private Credit going to be the straw that broke the Camel’s back\n\nWhat is Private Credit & Direct Lending\n\n2025E Private Lending ~2.7T & direct lending ~1.4T\n\nDirect Lending > 50% of all private credit\n\nEvolution of Direct Lending - Morgan Stanley – 2026Private credit investments: 'Some caution is reasonable,' advisor says\n\nPrivate credit is lending outside the traditional banking system - private lenders originate loans directly with borrowers\n\nPost-2008, banks retreated from lending to smaller firms. Direct lending filled that gap for middle market companies\n\nBanks are still exposed - they fund private lenders (BDCs)\n\nRetail/HNW investors exposed via BDCs, new private credit ETFs & Pension fund investments\n\n“A significant private credit shakeout on par with Covid losses is coming” – Morgan Stanley\n\nPrivate credit shakeout matching Covid losses coming, Morgan Stanley saysEvolution of Direct Lending - Morgan Stanley - 2026\n\n“In our view, AI disruption will be a meaningful catalyst to drive defaults higher in direct lending… Overall, we expect the direct lending default rates to reach 8%, approaching Covid peak levels”\nstrategist Morgan Stanley Joyce Jiang (as reported by CNBC)\n\nPremise: New technology & AI ➜ erode demand for software services ➜ hurt lenders\n\nMorgan Stanley estimates that ~26% of direct lender portfolios (BDCs) and ~19% of private credit CLOs are exposed to software - a sector highly vulnerable to AI disruption\n\nInvestors in private credit demanding their money back\n\nBlackstone's flagship private credit fund posts first monthly loss in over three years | ReutersWhy Wall Street is calling out ‘echoes’ of the 2008 financial crisis | CNN BusinessAfter pouring billions into private credit, many investors want out\n\nBlackstone, the private equity giant: $3.8 billion in redemption requests. At least 25 senior leaders from across the firm pitched in some $150 million from their own wallets .\n\nBCRED, Blackstones flagship private credit fund posted its first monthly loss of 0.4% in more than three years in February, it’s first since September 2022 ,when it posted a loss of 1.3%.\n\nBlue Owl - hit with withdrawal requests. Halts redemptions and liquidates assets to repay.\n\n“When you see one cockroach, there are probably more… Everyone should be forewarned on this” Jamie Dimon\n\nWhy Jamie Dimon is warning of ‘cockroaches’ in the US economy | CNN Business\n\nBeyond software, the other stress point is subprime auto lending. In 2008 it was toxic mortgages - today the red flag is car loans\nTricolor Holdings (subprime auto lender) went bust Sep 2025. Even JPMorgan took a $170M hit from this single bankruptcy\nFirst Brands (auto-parts) filed Chapter 11 with ~$2.3B in hidden loans. DOJ opened a criminal probe - echoes of Lehman\n\nDimon’s point: these may be isolated, but in a downturn more will surface - “when you see one cockroach, there are probably more”\n\nAre Car Loans the next ‘toxic mortgages’\n\nStructural weaknesses intensifying\n\n↓\n\nNon Farm\nPayroll\n\n2-yr downtrend\n\n↑\n\nJobless\nRate\n\nCreeping up\n\n↑\n\nBrent\nCrude\n\nNearly 2x\n\n↑\n\nPPI\nCore\n\n15-yr high\n\n↑\n\nPPI\nOverall\n\nOil shock pending\n\n↓\n\nRetail\nSales\n\nDeclining\n\n↑\n\n30+ DPD\nDelinquency\n\nPast COVID highs\n\n↓\n\nYield\nCurve\n\nFalling\n\n↓\n\nEquity\nMarkets\n\n5-10% off highs\n\nUnemployment creeping up steadily the past 2 years\n\nPast 20 Years\n\nBrent moving towards 2008 territory\n\nNearly doubled in past 3 months – supply shock as strait of Hormuz remains closed\n\nPast 25 Years\n\nUS 30+ DPD has already crossed the COVID highs\n\nPast 20 Years\n\nMarkets are already pricing in the weakness\n\nS&P 500 is down ~5% from its Jan 2026 high\n\nNifty 50 is down ~10% from its Jan 2026 high\n\nDAX is down ~8% from its Jan 2026 high\n\nNo predictions.\n\nJust data. Tracked daily\n\ntremor.tigzig.com\n\n10 Signals – all pointing in same direction\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: Quant apps hub, Analysis hub, APIs and MCP\n\nMore posts: The Crisis in Auto Loans. 2007 was Sub-Prime Mortgage. Is 2026 Auto?, Dimon Was Right. More Cockroaches Arrive. UK Lender MFS Collapses - $2.5B Across 12+ Global Banks., Private Credit - The $2.7 Trillion Shadow Lending Market Is Showing Cracks, The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top., Are We Headed for Stagflation-Lite"
},
{
"slug": "talk-to-your-database-from-excel-mcp-part-2",
"title": "Talk to Your Database from Excel via Claude & MCP - Part 2",
"tags": [
"claude-in-excel",
"mcp"
],
"content": "Talk to Your Database from Excel via Claude & MCP - Part 2\n\nClaude Excel MCP Database Part 2\n\nTwo new things.\nOpen public server - now upgraded. No API key, no login. Hardened with rate limits, concurrency caps, SQL validation etc. 23 defense layers total. Open for anyone to try - or pressure test. Have at it.\nOAuth-secured server (Auth0) - this is how I typically share tools with clients. Restricted access, JWT validation. If you want access to test, reach out. Once OAuth is at backend, setting up role-based access control is straightforward from there. The IT world has been doing this for decades - nothing new here.\n\nFull source code is open. Single Python file. Clone it, inspect it, deploy your own.\n\nMy standard MCP advice still stands - don't connect to unknown MCP servers blind. Use servers from reputed providers, or ones you deploy yourself.\n\nIf you'd like to implement, easiest way - paste this message to your AI Coder. It's AI-agent first site. It will fetch llms.txt and pull out everything you need.\n\nFull security checklist for Web apps: tigzig.com/security\n\nPart 1 - Talk to Your Database from Excel via MCP (with all links)\nhttps://www.tigzig.com/post/talk-to-your-database-from-excel-postgres-duckdb-claude-mcp\n\n\nRelated\n\nTools: Database Query Tool, DATS-4 Database AI Suite, BRIQ\n\nExplore: Database AI Apps, Claude for Analytics, API and MCP catalog\n\nMore posts: Talk to Your Database from Excel - Postgres, DuckDB - via Claude in Excel with MCP, Claude in Excel with MCP Connector - Talk to Your Backends from Inside Excel, MCP Guide for Claude in Excel - and any MCP client, Security Checklist for Web Apps - 71 Items"
},
{
"slug": "talk-to-your-database-from-excel-postgres-duckdb-claude-mcp",
"title": "Talk to Your Database from Excel - Postgres, DuckDB - via Claude in Excel with MCP",
"tags": [
"claude-in-excel",
"mcp",
"duckdb",
"database-ai"
],
"content": "Talk to Your Database from Excel - Postgres, DuckDB - via Claude in Excel with MCP\n\nI have put up a public MCP server that lets you run SQL queries against two live databases - right from Claude in Excel. No setup, no credentials, no installs. Just connect and ask.\n\nThe server sits on top of two cricket databases - about 2 million rows of ball-by-ball data. One on Postgres (Supabase) with ODI records. One on DuckDB with T20 records. Both covering 2013 to 2025.\n\nEach row is a single delivery in a match. So you have the batsman, bowler, runs off bat, extras (wides, no-balls, byes, legbyes), wicket type, player dismissed, venue, batting team, bowling team, innings, season, match date. Ball-by-ball level detail for every ODI and T20 match over 12 years.\n\nYou can ask things like \"top 20 strikers by total runs, matches, overs faced and run rate per match and per over\" Claude figures out the SQL, hits the database, brings back results right into Excel.\n\nYou open Claude in Excel, go to Settings > Connectors > Custom Connectors, paste the MCP URL, give it a name, save. Come back to the chat and start asking. That's it.\n\nI have also put up a sample spreadsheet with data pulled from this server and charts built on top of it - link in resources below.\n\nHow does the agent know what the data is\n\nThis is an important practical point. When Claude connects to the MCP server, how does it know what tables exist, what columns mean, what the data looks like?\n\nIn this case I have added a detailed docstring as part of the MCP tool description itself. So when Claude connects, it gets the table name, all column names, data semantics (what each field means), counting rules (the ball field is an over.ball identifier, not a sequential count - overs can have more than 6 deliveries due to wides and no-balls), how runs work (runsoffbat vs extras), how to check for wickets (both wickettype and otherwicket_type), player name search tips (use LIKE wildcards with surnames), season format quirks (can be '2023' or '2023/24'). Plus an example query to get it started.\n\nThis is enough for a single-table cricket dataset. The agent picks it up and writes correct SQL most of the time without needing additional guidance.\n\nBut in real life production setups - where you have 10s or 100s of tables with business logic, complex joins, calculated fields - this docstring approach is not enough. You need a proper semantic layer. That means business definitions for every table and column, relationship mappings between tables, golden queries for common business questions (pre-validated SQL that you know gives correct results), and rules for things like \"revenue means net revenue not gross\" or \"active customer means at least one transaction in last 90 days.\"\n\nThat semantic layer itself becomes a whole operation to build and maintain. How detailed should it be, how do you feed the right context for each query without bloating the prompt, how do you keep it updated as the schema changes. I have built this for client projects and it is real work.\n\nFor this public demo server though, the docstring approach works well. Simple schema, clear data semantics, and the agent gets it right.\n\nHow it works under the hood\n\nIt is a FastAPI server. For Postgres, the connection goes through asyncpg with a connection pool. For DuckDB, it runs through a read-only embedded connection. Both are exposed as API endpoints - /api/query/postgres and /api/query/duckdb. These endpoints are then mounted as an MCP server using Tadata's FastAPI-MCP package. So any MCP-compatible client can connect - Claude in Excel, Claude Code, Claude Desktop, or anything else that speaks MCP.\n\nThe server supports two response formats. JSON is the default. TSV is the compact option - uses shortened column headers and is about 70% smaller, which matters when you are working within AI context windows.\n\nWhere it stumbles and what I have done about it\n\nIf you have read my earlier post on Claude in Excel with MCP, you know the pattern. Claude in Excel has a tendency to pull everything at one shot and then try to write it all to Excel at once. It basically thinks it is Claude Code - dumps the entire result set into its context window, tries to write thousands of cells in one go, then chokes up and collapses.\n\nFor this database server I have handled it from the backend itself. Responses are capped at 1,000 rows. If your query returns more, it truncates to the first 1,000 rows and sends back a truncation flag so you know the result is not complete. It does not throw an error - you still get data, just capped.\n\nThe assumption is this: if you are querying from Claude in Excel, you are pulling summaries, aggregations, filtered record sets - not dumping a million rows into a spreadsheet. A query like \"top 20 run scorers by season\" returns 20 rows. \"Match summary for all T20s in 2024\" might return a few hundred. That is the sweet spot for this setup.\n\nIf you need bulk data extraction - full table dumps, millions of rows - this is not the right tool. Use Claude Code or a direct database connection for that.\n\nEven within the 1,000 row cap, if you are pulling 500+ records, you might want to tell Claude to write data in chunks rather than all at once. Something like \"pull the data and write it in batches of 100 rows.\" Otherwise it can still struggle with the writing-to-cells part. The data pull itself is fine - it is the writing to Excel where it gets clumsy with larger sets.\n\nFor most analytical queries though - top scorers, averages by team, match summaries, bowling figures, strike rates - results are typically a few dozen rows. That works cleanly with no issues.\n\nA standard test query\n\nHere is the query I use to test the setup:\n\n\"Show top 20 strikers by total runs off the bat, matches and overs played and run rate per match and per over. Share that as separate tables. Add a few insightful charts for each.\"\n\nThis is a good test because it requires the agent to write multiple SQL queries (aggregations, computed columns like run rate), create separate result tables in Excel, and then build charts. It exercises the full pipeline - understanding the data, writing correct SQL, presenting results, and visualization. Try this one when you connect - it works reliably.\n\nAvoid queries with specific player name searches as a first test. Name matching can be dicey - some players have initials, some have full first names, some have multiple spellings. Use the LIKE wildcard approach (WHERE striker LIKE '%Kohli%') rather than exact name matches until you know how names are stored.\n\nSecurity\n\nThis is a public server, kept open deliberately for people to try. So I have not put OAuth or role-based access control on it. But the safety basics are all there:\nThe Postgres user itself is read-only at the database level\nEvery connection sets defaulttransactionread_only = on\nDuckDB opens in read_only mode with external access disabled\nSQL validation blocks INSERT, DROP, ALTER, CREATE, COPY, LOAD, ATTACH and about 20 other dangerous keywords - checked as whole words to avoid false positives\nOnly SELECT, SHOW, DESCRIBE, EXPLAIN, WITH queries are allowed\nQuery timeouts on both databases (30 seconds default)\nRate limiting at 60 queries per hour per IP\nRow cap at 1,000 rows per query (truncated with flag, not errored)\n\nIf you need proper access control - MCP does support OAuth, and Claude's custom connector settings have fields for OAuth Client ID and Client Secret. So in theory you can set up proper authentication with an identity provider like Auth0, add role-based access, control who sees what data. I have not implemented it on this server because it is a public demo. And to be honest, I have not tested the OAuth flow end to end with Claude in Excel yet - there have been some reported issues with OAuth in custom connectors, so this is something I will test and cover separately. For my client work currently, I handle access control with token-based auth at the API level.\n\nOn MCP security - the usual reminder\n\nDon't connect to random MCP servers. You are basically running somebody else's code. You never know what an unknown MCP endpoint is doing with your queries or your data.\n\nThis server is fully documented with source code on GitHub. You can read every line of the Python code - it is a single file, about 550 lines. If you don't trust using my endpoint directly, clone the repo, inspect it, and deploy on your own infrastructure. Or just test it with the public endpoint first since it is read-only anyway.\n\nSource code and how to replicate\n\nThe full source code is on GitHub as a public repo. The README has setup instructions - clone, install, add your Postgres connection string, run. The repo includes the DuckDB file with T20 data ready to go.\n\nIt is a single Python file. FastAPI, asyncpg, duckdb, fastapi-mcp, slowapi for rate limiting. Clean and simple - you can read and understand the whole thing in one sitting.\n\nResources\nDocumentation page\nMCP Endpoint: https://db-mcp.tigzig.com/mcp\nAPI Docs (Swagger)\nSource Code on GitHub\nSample spreadsheet with charts\n\nPrevious posts on Claude in Excel and MCP:\nPower User Guide to Claude in Excel & PowerPoint - 26 Working Tips\nClaude in Excel with MCP Connector - Talk to Your Backends from Inside Excel\nClaude in Excel - Nifty50 Return Distribution Analysis\nClaude in Excel built a 50-chart India Macroeconomic Dashboard from RBI data\nClaude in Excel just one-shotted an XGBoost response model\n\nCricket data from Cricsheet, published under the Open Data Commons Attribution License 1.0 (ODC-BY).\n\n\nRelated\n\nTools: Database Query Tool, BRIQ, DATS-4 Database AI Suite\n\nExplore: Claude for Analytics hub, Database AI apps, MCP and API hub\n\nMore posts: Talk to Your Database from Excel via Claude & MCP - Part 2, Claude in Excel with MCP Connector - Talk to Your Backends from Inside Excel, MCP Guide for Claude in Excel - and any MCP client, My Public MCP Server Got Hammered - Security Lessons from a Bot Attack, Cricket (ODI/T20) & Tour de France stats from a Custom GPT connected to 3 live databases."
},
{
"slug": "that-9x-return-from-nifty-midcap-is-irrelevant-if-you-couldn-t-survive-the-73-of-time-it-was-in-dra",
"title": "That 9X return from Nifty Midcap is irrelevant if you couldn't survive the 73% of time it was in drawdown",
"tags": [
"portfolio-analytics"
],
"content": "That 9X return from Nifty Midcap is irrelevant if you couldn't survive the 73% of time it was in drawdown\n\nI analyzed 18 years of data for three iconic Indian investments: the Nifty 50, L&T, and the Nifty Midcap 100.\n\nThe results on paper are spectacular. But they hide a brutal psychological truth.\n\nI designed the Anxiety Index to quantify the psychological impact. An index of 73% means you spent 73% of your holding period watching your investment sit below a previous peak (exact definitions at the end)\nTo earn that 9X in midcaps, you had to endure a 73% Anxiety (Index)\nFor L&T's impressive 7X growth, the price was a 75% Anxiety (Index), including a single drawdown that lasted 2,377 days.\n\nLet that sink in. 6.5 years watching your investment be worth less than it was before. Even the 'safer' Nifty 50 spent 63% of the time in a drawdown.\n\nYour biggest risk isn't a number on a spreadsheet. It's your own behavior in the face of a prolonged, gut-wrenching drawdown.\n\nYour financial plan is irrelevant if you don't have the conviction to execute it when you are in depths of a bear market\n\nBefore you chase the next 10X idea, ask yourself:\nDo I have the temperament to go nowhere for 5+ years?\nIs my financial life structured to survive needing money at the worst possible time?\nIs my asset allocation built for my psychology, not just for a model?\n\nPast performance is no guarantee of future results. But it is a hell of a teacher about the price of returns.\n\nQuantify the psychological price of your own investments. I built TIGZIG Quants for this analysis - it's free. Analyze any stock, metal, or crypto across your chosen time window.\n\nP.S.\n\nDrawdown measures the fall from an asset's most recent peak. You can be profitable overall but still be in a 20% drawdown. Anxiety Index is my metric for quantifying the psychological cost of watching your portfolio drop from a peak and waiting for it to recover. It is calculated by summing the days of most significant drawdown periods (I take the worst seven) and expressing this as a percentage of the total time.\n\nThis is not investment advice. I'm a CA, data scientist, ex-trader and long time investor building tools for my own research. The code is open-source.\n\n\nRelated\n\nTools: QRep - Security Reports, Quants Agent, Quants Suite\n\nExplore: Quant Apps hub, Analysis hub, Guides hub\n\nMore posts: S&P 500 Drawdown Analysis with QREP, Gold up 2.3X past 2 years. But what about the 10 years drawdown in between., U.S. Markets (S&P 500) vs India (Nifty 50) - is the returns profile reversing?"
},
{
"slug": "the-google-machine-continues-to-roll-will-it-do-to-ai-what-it-did-to-search",
"title": "The Google Machine continues to roll. Will it do to AI what it did to search?",
"tags": [
"ai-coders"
],
"content": "The Google Machine continues to roll. Will it do to AI what it did to search?\n\nGoogle AI Tools\n\nA month ago, I posted that Google was back in the AI race. Earlier in the year, I wrote about the shift in their LLM quality.\n\nSanya Ojha's analysis this week confirms it from the strategy side. I'm seeing it from the execution side.\n\nGoogle is shipping at a rate I haven't seen in years. Around 20% of my workflow now runs on Gemini. In January 2025, that was 0%.\n\nHere's what I am using for my client projects\ngemini-2.5-flash-lite - my workhorse. first choice for automations. Fast, great quality, cheap.\nGoogle AI Studio (aistudio.google.com) - app builder for rapid prototypes before I move to Cursor or Claude Code. The interface improved massively. You can create a project right there without touching Google Cloud Console. Usage tracking is solid. Works for anyone - students, PMs, analysts - who need to prototype apps fast. And small apps - pretty much 90% can be built there.\nGemini CLI - AI coder with a huge free tier. Part of my primary stack now, alongside Claude Code and Cursor. It's still far away from Claude Sonnet 4.5 or GPT-5, but great from many lighter tasks. Evolving fast and will only get better.\nMCP for Chrome DevTools - powerful for reviewing browser console logs and letting an AI coder use the browser directly.\nMCP toolbox for databases - launched earlier this year as a middleware layer for connecting AI/non-AI frontends to backend databases. I use a simpler custom-built layer, but this provides a robust framework. Testing it soon.\nFile Search API - just launched. Supports the full RAG pipeline: vector database, embeddings, chunking. You upload files, Google handles the backend, you send user queries to the API. Very basic right now, limited config options, but my sense is it will become a full-fledged RAG toolbox. Pricing is excellent with a generous free tier.\n\nAt the rate they're going - straddling the whole food chain from LLMs to hardware, to builder tools & enterprise integrations - there's a fair chance in a few years they'll do to AI what they did to search.\n\n➜ But one big gap remains: Top tier LLM and AI coding. For high-intelligence work, Claude Sonnet 4.5 is still the leader, followed by OpenAI. Google sits at #3. They've won the workhorse tier - fast, cheap, reliable bulk work. They haven't won the frontier tier yet\n\nWill they get there? Based on the way it is gaining grounds, odds are in favor of Google.\n\nRead Sanya's full analysis\n\nMy original post from last month: Google - the old edge is back\nsource: migrated)\n\nGoogle AI Analysis\n\n\nRelated\n\nTools: Portfolio Analytics Agent, TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Build guides, Database AI Apps, Browse all apps\n\nMore posts: Google - The old edge is back.By Dec ’24, in AI, I had written Google off. Now, the balance has shifted, Google Tools I Use on Live Projects — Analysis, Automation & Building Micro-Apps, Google Antigravity just launched. For analysts and data scientists: Worth adding to your toolkit"
},
{
"slug": "the-xlwings-lite-ai-coder-instruction-file-december-2025-release",
"title": "The xlwings Lite AI Coder Instruction File - December 2025 Release",
"tags": [
"xlwings-lite",
"python-in-excel",
"ai-coders"
],
"content": "The xlwings Lite AI Coder Instruction File - December 2025 Release\n\nxlwings Lite\n\nxlwings Lite turns Excel into a live Python app. No Python install on the user's machine. Just a browser-based add-in with code editor, console, and deep Excel integration. It launched earlier this year and I've been using it ever since.\n\nWhen you use AI coders like Claude, Cursor, or Gemini to write xlwings Lite code, they don't know its constraints. They generate desktop xlwings syntax that looks right but fails silently. So I started maintaining an instruction file - a document you feed to the AI before it writes code, teaching it what works and what doesn't in xlwings Lite's environment.\n\nStarted as a handful of rules. Went to 1,450 lines by July. Now at 1,867 lines. The additions are patterns I had to figure out over five months of client work and tool builds.\n\nLatest client build using this file: A data cleaning workflow for a team of 4-5 domain experts. They connect to a read-only database view, filter by date range and segment, pull records. An LLM module runs and shares suggestions for each record. The domain folks review, tag off wherever the AI suggestions are acceptable, adjust where needed. The script generates a CSV that goes to the data guy for final validation before the actual backend update. Users never touch the live database directly - read-only access, manual validation step, then the data guy does the insert. Built the whole thing in xlwings Lite with the instruction file guiding the AI coder. Took a fraction of the time it would have otherwise.\n\nFor more open-source builds and public tools, see the sections below.\n\nWhat's in the December 2025 Version\n\n21 Golden Rules (was 5 when I started off)\n\nThese are the non-negotiable patterns. Violate any of them, and your script either fails or produces garbage. Examples:\nNever use .expand() on data you just wrote. Excel hasn't registered it yet. Use .resize() with explicit dimensions instead.\nNever write 2D Python lists to a single cell. Causes silent InvalidArgument errors. Write cell-by-cell.\nNever mix numbers and empty strings in a DataFrame column. Use None for missing numeric values, or convert the whole column to string.\nNever access a Table's parent sheet with .sheet - the attribute doesn't exist in xlwings Lite. Use the helper function that searches all sheets.\n\nEach rule exists because I hit the problem on a live project. Documented the failure mode. Added the fix.\n\nInvalidArgument Troubleshooting Guide\n\nThis error shows up in the Excel UI with zero Python traceback. Top three causes, in order of likelihood:\nMixed data types in a DataFrame column (most common - the number vs empty string problem)\nSheet state contamination from manual renaming (user renamed an output sheet, API gets confused)\nGeneral API instability requiring a forced roundtrip before table creation\n\nThe guide walks through diagnosis and fixes for each. Before this section existed, I had to constantly look into the codes and keep reminding AI.\n\nCustom Function Patterns\n\nThe @func decorator for Excel custom functions has a gotcha: if your type hint says float and the Excel cell contains text or is blank, you get #VALUE!. The fix is using typing.Any and converting inside the function.\n\nAPI Stability Workarounds\n\nxlwings Lite talks to Excel through a JavaScript bridge. Sometimes the API state gets out of sync. The 'API Sync' pattern forces a roundtrip before critical operations - write a dummy value, read it back, then proceed. A bit messy, but it works.\n\nxlwings Lite Limitations\n\nThe instruction file has the complete list. When your AI Coder knows the boundaries upfront, you design around them instead of discovering them mid-project.\nsheet.autofit() - NotImplementedError\nRange.current_region - Not implemented\nRange.merge() / Range.unmerge() - Not implemented\nNo direct database connections (need a web API layer)\nNo multiprocessing or threading\nbook.save() doesn't exist\n\nThe 4 System Constraints\n\nHighest priority. AI must verify against these before outputting code. These four will save you and your AI Coder the most debugging time per line of instruction.\nNo indexes when writing DataFrames - Always .options(index=False)\nRobust sheet deletion - Check existence before deleting, create fresh\nNo autofit - Use explicit column widths or defaults\nNo RGB tuples for colors - Use hex strings like '#F0F0F0'\n\nWhat I've Built With This\n\nWorking apps, all available with files and docs at the xlwings Lite Practice Lab xlwings-lite.tigzig.com\nAI Web Scraper - Paste URLs, define extraction rules in plain English. Jina AI fetches pages, Gemini extracts structured data into Excel tables.\nAI Technical Analyst - Yahoo Finance data to charts to Gemini Vision analysis to PDF reports. End-to-end.\nMF Holdings Analyzer - Processes monthly mutual fund portfolio disclosures (India market). Standardizes names, merges ISINs, flags data issues for human review, outputs consolidated summary with charts. Two-stage workflow with human-in-the-loop quality control.\nDatabase Connector - Connect to remote databases through a FastAPI layer. Browse schema, run queries, all from Excel.\nEDA + ML Workflow - Dynamic plots, summary tables, XGBoost model with decile analysis and gains charts.\n\nHow to Use the Instruction File\nDownload from the xlwings Lite Practice Lab at www.tigzig.com\nUpload to your AI coder - Cursor, Claude Code, Gemini CLI, Google Antigravity.\nGoogle AI Studio with Gemini 2.5 Pro works well (free tier, 1M context) for quick uses.\n\nThe file has two versions: AI-optimized (for machine consumption) and Human-readable (same content, better formatting for reference).\n\nDo You Need to Know Python?\n\nIt helps. But you can start without it. The actual skill isn't writing code. It's designing the analytics, spotting data problems, validating results, and making it work in the real world. That's the expertise AI can't replace. If you're an analyst who's never touched Python, you'll learn what you need by doing. The instruction file handles the xlwings-specific gotchas. Your domain knowledge handles the rest.\n\nThe 5 Hard Rules for AI-Assisted Coding\n\nSame rules I follow for any AI coder work, not just xlwings:\nBe specific - Vague instructions produce vague code\nIterate - One step at a time. Validate. Debug.\nReview - Demand a plan and pseudocode. Interrogate AI\nValidate - Test it. Break it. Fix it.\nAI Audit - Final pass in a fresh session catches what you missed\n\nResources\nAI Coder Instructions file, live tools and guides at xlwings Lite Practice Lab - xlwings-lite.tigzig.com\nxlwings Lite Official Docs: lite.xlwings.org\n\nxlwings Lite official site: lite.xlwings.org\n\n\nRelated\n\nTools: AI Web Scraper, Technical Analysis in Excel, MF Holdings Analyzer\n\nExplore: xlwings Lite Practice Lab, AI in Excel hub, Database AI Apps\n\nMore posts: Tool: A 1,450-line context file. Purpose: To ensure clean, efficient xlwings Lite code generation., Which AI Coder should you use for xlwings Lite (Python in Excel)?, Python In Excel - Claude Vs. xlwings Lite? Who Wins?, Python in Excel: Field Guide & Practice Lab for AI-assisted xlwings Lite."
},
{
"slug": "think-about-it-one-of-the-world-s-top-ai-researchers-is-building-tools-deploying-them-live",
"title": "Think about it. One of the world's top AI researchers is building tools. Deploying them live.",
"tags": [
"ai-coders"
],
"content": "Think about it. One of the world's top AI researchers is building tools. Deploying them live.\n\nAndrew Ng released an 'Agentic Reviewer' for research papers. Started it as a weekend project.\n\nThink about it. One of the world's top AI researchers is building tools. Deploying them live.\n\nThat report you keep generating manually. That dashboard someone keeps asking for. That AI app idea you've been 'meaning to build'. Why aren't you?\n\nThe barrier is gone. AI coders (Claude Code, Cursor) turned domain experts into builders. In Andrew's own words \"I've long been fan of Claude Code\"...\n\nOver the past two years, I've built a Micro-apps portal with 30+ apps at www.tigzig.com with AI Coders. Database AI, quants tools, automations. Currently working on a 20-user internal AI tool with React + PHP. Two years ago, I'd have passed on that project.\n\nWhere to start\n\nPay $20/month for Claude Code (my top choice) or Cursor. From your pocket if needed. Want to test free first? Google Antigravity (antigravity.google) - one-click install, small message limit but enough to get a feel for it. Then move to Claude Code or Cursor for real work. Gemini CLI (free) for lighter tasks. Mito AI for notebooks. Great resources available for free on YouTube from basic apps to complex full stack builds (links below)\n\nAndrew Ng's post\n\nhttps://lnkd.in/gBGJmeyp\n\nYouTube guides\nVolo Builds: youtube.com/@VoloBuilds\nLeon Van Zyl: youtube.com/@leonvanzyl\nMark Kashef: www.youtube.com/@Mark\\_Kashef/videos\n\nMy posts on AI coders\nThe 6 rules I follow\nCoding by hand is obsolete\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, BRIQ\n\nExplore: Browse all apps, How I work, APIs and MCP\n\nMore posts: Andrew Ng is using Claude Code, OpenAI Codex, Gemini CLI., Google Antigravity just launched. For analysts and data scientists: Worth adding to your toolkit, AI Coders are here. The edge now is domain + execution. Not vibing."
},
{
"slug": "three-red-flags-30yr-ppi-oil-may2026",
"title": "Three Red Flags Hit Together. 30Y at 5.14%, PPI at 5.99%, Brent Above $100. Setup Worse Than 2000 or 2007.",
"tags": [
"portfolio-quants"
],
"content": "Three Red Flags Hit Together. 30Y at 5.14%, PPI at 5.99%, Brent Above $100. Setup Worse Than 2000 or 2007.\n\nPublished: 2026-05-20\n\nThree macro red flags hit at the same time, each a major disruptor on its own: US 30-year Treasury yield 5.14% (first since 2007), US Producer prices 5.99% (15-year high ex-COVID), Brent above $100 (IEA calls it the largest oil supply disruption in history). Alongside: S&P 7,400 with CAPE 39.6 (2nd-highest in 145 years), auto loan losses past 2007 GFC peak, unemployment near Q3 2007 levels, Mag 7 ~35% of S&P (NVIDIA alone 8.9%). The last time these general patterns showed up: 2000 and 2007. 2026 is looking worse.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/three-red-flags-30yr-ppi-oil-may2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/three-red-flags-30yr-ppi-oil-may2026 with header Accept: text/markdown\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nMacro Stress Note · 20 May 2026 · Amar Harolikar · Decision Sciences & Applied AI\n\nThe bond market, the producer, and the oil tanker — all three pointing the same way.\nUS 30-Year Treasury Yield\n\nBack at 5.14%. First time at 5% since 2007.\n\nThe Fed is boxed in.\nBond vigilantes pricing 5%+.\nNew Treasury issuance hitting weak auctions.\nMarkets price 56% odds of a December hike.\nJGB yields at multi-decade highs may pull Japanese capital out of US Treasuries.\n\nChart range: Jan 2020 — May 2026, YoY %.\nUS PPI Final Demand\n\nAt 5.99% in April. Highest since 2011, excluding the COVID spike.\n\nThe US cascade.\nCost of capital up across the economy.\nAI capex will likely face a squeeze.\nConsumer delinquencies at 3.4% — past the 2007 pre-crisis level of 3.1%.\nEquities pinched both ways: higher discount rate + earnings uncertainty.\n\nChart range: Jan 2020 — Apr 2026, YoY %.\nBrent Crude Oil\n\nAbove $100 sustained. IEA calls it the largest oil supply disruption in history.\n\nWhat this means for India.\nInflation rising — LPG and diesel hit first.\nIT layoffs; tech valuations down.\nNo more the EM darling.\nBrokerages cutting India GDP and earnings forecasts.\n\nChart range: Jan 2020 — May 2026, USD / bbl.\n\nSources\nData: tremor.tigzig.com — FRED, BLS, US Treasury, NY Fed\nQuote: IEA Oil Market Report, March 2026\nCoverage: CNBC, Fortune, Yahoo Finance\n\nRun your own charts, cuts and analytics: tigzig.com → TREMOR\n\nPublished: 20 May 2026.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QRep - Security Reports, Quants Agent\n\nExplore: Market tools hub, Analysis deck, Open data APIs and MCP\n\nMore posts: S&P at 7,400. Irrational Exuberance. The Last Two Times This Setup Held, the Index Halved., The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top., The Masking Effect in Oil Markets - The Price You See on Screens Is Hiding a $50 Supply Shock, Which Chart Is the Odd One Out? Three Macro Charts Tell the Same Story. One Doesn't., The Crisis in Auto Loans. 2007 was Sub-Prime Mortgage. Is 2026 Auto?"
},
{
"slug": "three-tools-market-analysis-qsuite-tremor-qrep",
"title": "Three Tools I Built for Market Analysis - QSUITE, TREMOR, QREP",
"tags": [
"portfolio-quants"
],
"content": "Three Tools I Built for Market Analysis - QSUITE, TREMOR, QREP\n\nPublished: 2026-03-29\n\nThree market analysis tools - QSUITE for AI technical analysis, TREMOR for macro early warning with 50+ indicators across US and India, QREP for portfolio stats powered by QuantStats. Built on Claude Code, DuckDB, Hetzner, Vercel, Coolify. Free, no login, no ads.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/three-tools-market-analysis-qsuite-tremor-qrep\nMarkdown of the HTML page: send GET https://tigzig.com/post/three-tools-market-analysis-qsuite-tremor-qrep with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/QUANT_TOOLS.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nThe 3 Tools I built and use\n\nAmar Harolikar. Decision Sciences & Applied AI TigZig.com\n\nto track and analyze marketsFree. No Login. No Ads\nQREP: qrep.tigzig.comPortfolio stats, drawdowns\nQSUITE: qsuite.tigzig.comAI Technical Analysis\nTREMOR: tremor.tigzig.comMacro Indicators\n\nBuilt by Analyst - For Analysts\n\nQSUITE : Quants Suite\n\nAmar Harolikar. Decision Sciences & Applied AI TigZig.com\n\nAI Technical Analysis\n\nType a ticker get AI technical analysis report in PDF & HTML with technical charts\n\nChoice of 9 LLMs including – Claude 4.6 and GPT 5.4\n\nTREMOR : Macro Indicators\n\nAmar Harolikar. Decision Sciences & Applied AI TigZig.com\n\nEquities, bonds, commodities, valuations\n\nInflation, unemployment, GDP, sentiments\n\nNews across 25+ feeds across 6 categories\n\nQREP : Portfolio Stats & Drawdowns\n\nAmar Harolikar. Decision Sciences & Applied AI TigZig.com\n\nMulti security comparisons - Risk, returns, drawdowns, benchmarks & tear sheets\n\nPowered by QuantStats\n\nAre We Headed for Stagflation-LiteAre we staring into face of a full bear market?US Banks & Non-Bank Lending - How Deep is the Exposure? Is there a Systemic Risk?Structural Weakness IntensifyingPrivate Credit - The $2.7 Trillion Shadow Lending Market Is Showing Cracks\n\nRecent Analysis : Available on Blog\n\nblog.tigzig.com\n\nAmar Harolikar. Decision Sciences & Applied AI TigZig.com\n\nTIGZIG : AI for Analytics\n\nTigZig.com\n\n40+ live tools. Free. Built by Analyst – For Analysts\n\nAmar Harolikar. Decision Sciences & Applied AI TigZig.com\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QRep - Security Reports, Quants Suite\n\nExplore: Market tools hub, Analysis deck, Guides and blog\n\nMore posts: Tremor - Macro Early Warning Signals - March 2026, Are We Headed for Stagflation-Lite, Structural Weakness Intensifying, US Banks & Non-Bank Lending - How Deep is the Exposure? Is there a Systemic Risk?, Private Credit - The $2.7 Trillion Shadow Lending Market Is Showing Cracks"
},
{
"slug": "tigzig-agent-data-surface-aug2026",
"title": "Point Your AI Agent at api.tigzig.com and It Works Out Which Dataset It Needs. 38 Million Rows Across India and the US, Free.",
"tags": [
"mutual-funds",
"vigil",
"portfolio-quants"
],
"content": "Point Your AI Agent at api.tigzig.com and It Works Out Which Dataset It Needs. 38 Million Rows Across India and the US, Free.\n\nTell your AI agent to go to api.tigzig.com and pull the full mutual fund NAV history since 2006, or India credit ratings, or US consumer delinquency, and it will work out which dataset it needs and download the file on its own. Close to 3.8 Cr (38M) rows across India and US, curated and updated daily. Free, no signup and no key.\n\nTwenty years of mutual fund NAVs, around 330 macro, credit and valuation indicators, and seven separate filing datasets for Indian listed companies covering ratings, insider trades, pledges and takeovers.\n\nJust ask your AI Agent:\n\n``\ngo to api.tigzig.com and ....\ndownload the full mf dataset for me\npull ratings and surveillance data for India\npull shiller index, US consumer delinquency and unemployment\n`\n\nThere are three ways to use it.\nbrowse and filter in the app if you are working by hand,\ncall the API if you are writing code,\nor take a whole table as a single file in CSV, Parquet, SQLite or DuckDB if you would rather work offline.\n\nFor humans:\nIndia MF: tigzig.com/mfpro\nGlobal Macros: tigzig.com/tremor\nIndia Signals: tigzig.com/vigil\nAPI Page: tigzig.com/apis\n\nWhat sits on the macro side and the filings side\n\nThe macro side carries around 330 indicators across eight groups, US macro, US credit stress, US valuation, US insurance, India macro, global macro, markets and currencies, with each indicator carrying its own notes on annualisation and sign conventions, because those are the things that ruin a comparison without showing up as an error.\n\nThe India filings side is seven separate datasets: credit ratings, insider trading, promoter pledge, encumbrance, takeover and SAST filings, surveillance flags, and related party transactions. Each one downloads on its own or you can take all seven as a single SQLite file.\n\nEvery table has a downloads manifest with row counts and a generated_at` stamp, so a script can check whether anything actually changed before pulling a file again.\n\nSome practical notes from using this myself\nRenamed mutual fund schemes still resolve. If you are holding an old ISIN from a scheme that has since been renamed or merged, it still returns the full NAV series rather than an empty result, which saves a fair amount of reconciliation work on older holdings.\nSIF is deliberately kept out of the NAV data and sits on its own at sif.tigzig.com.\nIf you just want to look up or check an ISIN, that is at isin.tigzig.com.\nThe filings update daily. The credit ratings run picked up CRISIL moving IndusInd Bank from a negative outlook to stable, filed on the 19th, and the surveillance snapshot is from yesterday evening.\nOn surveillance, 666 companies currently carry a flag of some kind. No Nifty 50 company is on that list today, but Hindustan Zinc is, on the encumbrance criterion, more than half the promoter holding pledged or otherwise encumbered. Ola Electric is there too, on long term ASM and as loss making.\nRelated party transactions is the biggest of the filings tables. A typical company files under 200 rows for the period and the largest file more than 6,000, which tracks how many legal entities a group runs.\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, TREMOR - Macro Stress Signals, VIGIL - India Red Flag Events\n\nExplore: API and MCP catalog, India red flags hub, Claude for Analytics hub\n\nMore posts: Pull India Mutual Fund NAVs Straight Into Excel. 50, 100 or More Schemes, in a Few Clicks., Related Party Transactions for Around 750 Listed Indian Companies Are Now on VIGIL, With the March Half-Yearly Filings In., MF NAV API FAQ. Why Am I Getting an Error, Why Is There No SIF NAV, and Why Does the AUM Look Wrong?, Tigzig is Built AI-Agent First. 300+ Assets, and the Fastest Way In Is to Ask Your AI."
},
{
"slug": "tigzig-agent-data-surface-oct2026",
"title": "Explore the Tigzig Data Surface With Your AI: What Is New in October 2026",
"tags": [
"database-ai",
"mutual-funds",
"vigil",
"portfolio-quants"
],
"content": "Explore the Tigzig Data Surface With Your AI: What Is New in October 2026\n\nTo explore the Tigzig data surface with your AI, just paste this into your agent: \"Go to api.tigzig.com. Tell me what's there and how we can pull it.\"\n\nWhat's new\nMutual funds: scheme details. 60 fields per scheme, covering fund size, fund manager, TER and BER, benchmark and riskometer, for about 2,100 funds across 8,600+ plans.\nGlobal macro: 26 new indicators covering US jobs and wages, inflation, Japan's 10-year bond yield, India IIP and WPI on the new base, and the Buffett indicator.\nCricket: 29 more SQL functions on the query endpoints, so more of the stats work happens inside the query itself.\nIndia company signals: related party transactions now cover FY26.\n\nUsage has broadened over the past month, with about a quarter more callers than the month before. People come in through Google Sheets and Apps Script, Excel and Power BI templates, n8n, Zoho and Power Automate workflows, Python and Node scripts....and AI agents including Claude, ChatGPT, Codex, Mistral and Grok. They feed portfolio trackers, market dashboards, financial planners, family wealth dashboards, DCF valuation tools, fund screeners and their own research pipelines.\n\nAsk for one item, a batch, or take the whole table as a file and work on your own machine.\n\nFor you: https://www.tigzig.com/apis\n\nFor your agent: https://api.tigzig.com\n\nSome quick pointers\nMutual fund scheme details: fund size, fund manager, benchmark and the scheme terms belong to the fund, so they repeat on each of its plans. To total fund size, count only the rows where funddedupflag is true.\nFund size and AAUM are different numbers. Fund size is the whole fund on one day, from AMFI's daily fund performance data. AAUM is the quarterly average for each plan.\nTER is matched only on the exact SEBI scheme code in the fund's own document. Where that code is missing or shared with another fund, TER is left blank.\nUS jobs and prices: thirteen US series have no value for October 2025, the month of the government shutdown, ten from the jobs survey and three CPI series. FRED publishes that month as blank, so a monthly change across it skips a month.\nIndia IIP and WPI on the 2022-23 base are separate series from the old-base ones. They remain separate.\nCricket SQL: the 29 new functions include percentiles, standard deviation, mode and rankings such as quartiles. A few, like median, are DuckDB only.\nRelated party transactions: half-years ending March 2025, September 2025 and March 2026.\n\nUseful posts to debug your API calls\n\nSome useful posts for Tigzig data surface users.. to debug their API calls .. learnings from live user logs\nA practical guide to Tigzig API errors, with tips, for MFPRO, TREMOR and VIGIL users.\nhttps://www.tigzig.com/post/tigzig-api-errors-practical-guide-jul2026\nWhen a Tigzig API returns an error, the response body usually has a hint.\nhttps://www.tigzig.com/post/api-error-hints-aug2026\nMF NAV API FAQ: why am I getting an error, and other common questions.\nhttps://www.tigzig.com/post/mf-nav-api-faq-aug2026\nCricket database: 14 things that trip people up, and the most common SQL error.\nhttps://www.tigzig.com/post/cricket-database-14-things-that-trip-people-up-sep2026\nhttps://www.tigzig.com/post/db-mcp-url-encoding-errors-aug2026\n\nThe Tigzig data surface for your AI agent, October 2026\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, TREMOR - Macro Stress Signals, VIGIL - India Red Flag Events\n\nExplore: APIs and MCP hub, What is new, Mutual funds hub\n\nMore posts: Point Your AI Agent at api.tigzig.com. 40 Million Rows of Public Data, and Half the Bytes Are One File, Scheme Details for Indian Mutual Funds Are Live on Tigzig: Fund Size, Fund Manager, TER and BER, Benchmark and Riskometer, Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users, MF NAV API FAQ. Why Am I Getting an Error, Why Is There No SIF NAV, and Why Does the AUM Look Wrong?, For Analysts Using the Tigzig Cricket Database: 14 Things That Trip People Up"
},
{
"slug": "tigzig-agent-data-surface-sep2026",
"title": "Point Your AI Agent at api.tigzig.com. 40 Million Rows of Public Data, and Half the Bytes Are One File",
"tags": [
"database-ai",
"mutual-funds",
"vigil",
"portfolio-quants"
],
"content": "Point Your AI Agent at api.tigzig.com. 40 Million Rows of Public Data, and Half the Bytes Are One File\n\nPoint your AI agent at api.tigzig.com and it works out which dataset it needs and pulls the file itself. Over 40 million rows across Indian mutual funds, cricket, global macro and India company filings. Refreshed daily. Free, no signup, no key.\n\nCricket is the addition since I last wrote about this. Every delivery in Tests, ODIs, T20s and the IPL, men's and women's, around 5 million balls from December 2001 onward, with the full match and player tables beside it.\n\nOver the last seven days the surface served more than 110,000 API calls and more than 1,700 file downloads, about 40 GB in all.\n\nOne file accounts for more than half of those bytes. The complete mutual fund NAV history in Parquet was pulled 135 times, and at 178 MB a copy that single file is over 23 GB of the total. People are taking the entire database in one go and querying it on their own machine.\n\nThat is the part I did not expect when I started publishing these files. I assumed the API would carry the load and the downloads would be occasional. The bytes say the opposite.\n\nFour ways in. Call the REST API. Point an MCP client at it. Run SQL against the cricket endpoint. Or take a whole table as CSV, Parquet, SQLite or DuckDB and work offline.\n\nFor you: tigzig.com/apis\n\nFor your agent: api.tigzig.com\n\nWhat sits behind those four names\n\nMutual fund NAVs, every Indian scheme AMFI has published since April 2006, about 3.74 crore rows including matured and merged schemes.\n\nCricket, ball by ball, every delivery in men's and women's internationals plus the IPL, with match info, playing elevens and a person registry.\n\nMacro, credit and markets, 300+ indicators covering US and India macro, credit stress, valuation and currencies, each carrying its own notes on annualisation and sign conventions.\n\nIndia company signals, the filings side, credit ratings, insider trades, promoter pledge, encumbrance, takeover and SAST, surveillance flags and related party transactions.\n\nEvery table has a downloads manifest with row counts and a generated_at stamp, so a script can check whether anything changed before pulling again.\n\nIf you are going to take the files\n\nTwo practical notes for anyone pulling the bulk files.\n\nRead the manifest first. It carries the row count, the byte size and a generated_at stamp for every file, so your job can decide whether to download at all. The files also honour If-None-Match and If-Modified-Since and return 304 when nothing has changed.\n\nPick the format for the job. Parquet is the smallest and DuckDB and Polars can query it remotely with range requests, so you can pull a few columns without the whole file. SQLite and DuckDB come as a single database if you want every table in one place. CSV is there when something downstream insists on it.\n\nThe refresh runs every day, so a daily job and one file gets you the same data a few thousand API calls would."
},
{
"slug": "tigzig-agent-first-ask-your-ai-aug2026",
"title": "Tigzig is Built AI-Agent First. 300+ Assets, and the Fastest Way In Is to Ask Your AI.",
"tags": [
"ai-coders",
"infrastructure"
],
"content": "Tigzig is Built AI-Agent First. 300+ Assets, and the Fastest Way In Is to Ask Your AI.\n\nTigzig is built AI agent first and a fair bit has moved in the past month. You are welcome to browse, but more efficient would be to ask ChatGPT or whichever AI you use ..\n\nIf there is anything specific you are looking for then just ask ChatGPT / Claude, Gemini, Meta AI .... something like\n\"i want to build excel tool... what does tigzig.com have?\"\n\"does tigzig.com have any analysis on Donald Trump's trades\"\n\"what's amar harolikar's view on economy and markets - pull out his posts from tigzig.com\"\n\nAll 300+ assets on the site, guides, live apps, repos, open APIs and MCP servers, are organized to be machine readable / Agent accessible.\n\nAnd if you are using an AI Agent / AI Coder (Claude Cowork / ChatGPT Work / Claude Code / Codex), they would be able to pull out the codes, API's and implement, deploy and run it for you ...\n\nRate limits, and what changed in the traffic\n\nAnd traffic pattern has changed too. A fair portion of calls now arrive from agents and automated tools, so some callers are running into rate limits they would not have hit earlier.\n\nThe rate limits are under active adjustment and you should start to see most of your calls go through.\n\nThese would also be documented across all surfaces .. response headers / 429's / machine readable roots and also the human docs. MFPRO API does most of it .. would be extended to other high usage APIs including YFIN & FFN.\n\nIf still facing issue, then do consider batching your requests as many API's support multiple tickers / codes ... or for TREMOR and MF API .. you can simply download the whole database .. manually or via API.\n\nAPI hub: tigzig.com -> API & MCP\n\nWhere the traffic is coming from now\n\nAnd by the way I am getting more and more requests via ChatGPT and Claude.\n\nMore people landing on my website via ChatGPT .. after Google that's the number 1 referrer.\n\nAnd API usage show regular ChatGPT (mostly must be Codex or Work), Claude (must be either CoWork or Codex)... as they send out standard headers and behave in a very predictable and systematic manner .. what i call 'AI agent like'.\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: API + MCP hub, Browse all apps, How I work\n\nMore posts: MFPRO Scheme Master: Every India Mutual Fund Scheme in One Small File, Free, MFPRO Mutual Fund API: Search Now Filters Across 12 Fields on a Scheme, Your Agents and Scripts Have Been Talking to Me. Here Is That Conversation in Plain English., Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users, TigZig is Built Agent First. 300+ Assets on AI for Analytics. Just Ask Your ChatGPT, Gemini or Claude."
},
{
"slug": "tigzig-agent-first-ask-your-ai-jul2026",
"title": "TigZig is Built Agent First. 300+ Assets on AI for Analytics. Just Ask Your ChatGPT, Gemini or Claude.",
"tags": [
"ai-coders"
],
"content": "TigZig is Built Agent First. 300+ Assets on AI for Analytics. Just Ask Your ChatGPT, Gemini or Claude.\n\nTigZig is built agent first. Live tools, guides, repos, on AI for analytics. Over 300 assets. Welcome to browse, but likely to lose your way and miss something. Best way - just ask your regular ChatGPT or Gemini or Claude or whichever AI you keep open.\n\"check tigzig for anything on market valuations\"\n\"what does tigzig have on database ai\"\n\"any good tool on tigzig for macro data?\"\n\"anything on app security on tigzig?\"\n\nor just simply\n\"whats tigzig about?\"\n\nAnd if you are using an AI Coder like Claude Code / Cursor / Codex, then:\n\"i want to connect my duckdb database to Claude in Excel and other AI agents .. with auth and all ... saw some open source things on tigzig .. can you assess cloning from there and deployment\"\n\nQuick links to a few hubs on tigzig.com\nMarket Tools: https://www.tigzig.com/markets\nAI Workers for Analytics: https://www.tigzig.com/claude-for-analytics\nIndia Mutual Fund Analytics - Tools, Data, APIs: https://www.tigzig.com/mutual-funds\nConnect AI Agents / Claude in Excel to your databases: https://www.tigzig.com/mcp-landing\nLatest analysis on India and global markets: https://www.tigzig.com/analysis\n\nChat apps vs Cowork apps vs Embedded apps vs AI Coders - what's the difference?\n\nWondering what the difference is between chat apps like ChatGPT, Claude AI, Gemini vs. coworking apps like Claude Cowork / ChatGPT Work vs. embedded apps like Claude in Excel vs. AI Coders like Claude Code / Codex?\n\nRead here:\nJamie Dimon Uses Claude Code to Generate Dashboards. Are You? You Need an AI Worker or AI Coder, Not an AI Chatter.\nAI Is Not Just Chat. Cowork Apps Are Where the Action Happens.\n\nStatus: observational. SIF is a new category and this collection is part of the beta testing of our data pipeline. I am running it on its own for a while so the issues can surface and be worked through before it goes into the mutual fund pipeline behind MFPRO. There is no uptime commitment and the shape of the files may still change, so if you are building something on it that you rely on, keep your own copy.
\nIt is a third-party copy of AMFI data and is not affiliated with or endorsed by AMFI.
\n![\"TigZig](\"/images/blog/tigzigguidesaskyourai_onepager.png\")
\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Browse all apps, Claude for Analytics hub, Analysis hub\n\nMore posts: Tremor: A Free, No-Auth API for 330+ Macro and Market Indicators, 350K records, Refreshed 3x/Day. Ask your AI to explore., Claude for Analytics on Tigzig.com - New Hub: Excel, PowerPoint, Cowork, 7 MCP Servers, 20+ Guides., New MCP Hub on Tigzig.com - 7 MCP Servers, 6 Live Apps, 16 Guides, 108-Point Security Checklist. Just Ask Your AI."
},
{
"slug": "tigzig-agent-first-just-point-your-claude-jun2026",
"title": "The Fastest Way to Use Tigzig.com? Don't Click. Point Your Claude / AI Agent at It.",
"tags": [
"portfolio-quants",
"mcp"
],
"content": "The Fastest Way to Use Tigzig.com? Don't Click. Point Your Claude / AI Agent at It.\n\nPublished: 2026-06-13\n\nThe fastest way to use tigzig.com is to not click - point your Claude or AI agent at it. The site is built AI-agent-ready: 40+ analytics apps, open analytics APIs, MCP servers, 200+ guides, plus a repos shelf - all free, no sign-up. An agent pointed at the site can find the right tool / guide / dataset for the task, read app and database data and run the analysis, call the analytics APIs, set up MCP servers, and even download / install / deploy an open-source repo. Claude Cowork is particularly good at this - it runs its own sandbox inside the Claude desktop app, so once a domain is enabled it makes the API calls itself, pulls the data, runs the analysis and hands back a PDF. Codex, Cursor, any agent works the same way. Rather explore yourself? The whole site is still right there - the agent just gets you there faster. Don't believe it? Point your agent at tigzig.com, let it explore, then ask it 'how do you like the site?'\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/tigzig-agent-first-just-point-your-claude-jun2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/tigzig-agent-first-just-point-your-claude-jun2026 with header Accept: text/markdown\n\nTags\nportfolio-quants, mcp\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: API + MCP catalog, Browse all apps, Market tools hub\n\nMore posts: AI Is the Front Door Now. Don't Dig Through Tigzig.com - Ask Your AI About It., Claude Cowork and Claude Code Have Quietly Become My Universal Front End. More Than Excel, Jupyter or Colab., TigZig is Now AI-Agent First"
},
{
"slug": "tigzig-ai-agent-first-site",
"title": "tigzig.com is AI-agent first. But what happens when your AI coder runs into a problem on my site?",
"tags": [
"ai-coders",
"infrastructure"
],
"content": "tigzig.com is AI-agent first. But what happens when your AI coder runs into a problem on my site?\n\nTigZig AI Agent Feedback\n\ntigzig.com is AI-agent first. But what happens when your AI coder runs into a problem on my site?\n\nSimple. Ask your agent to send me a feedback. It knows how - there's an API endpoint, your agent will figure out what to send and where to hit it.\n\nWhat happens next.. it lands in my ops monitoring center. My Claude picks it up, drafts a resolution based on my instructions, I approve... matter over.\n\nAdd your email if you want a reply - it'll come back to you.\n\nSo basically.. my AI talking to your AI.\n\nI think this reduces a lot of friction. No worrying about tone, tenor, phrasing.. two AIs pretty much stick to facts, execute based on our instructions ... and get along really well with each other.\n\nNo unnecessary human issues to tackle ...they just get stuff done.\n\nLet them do what they do best.. so we can focus on the business of making money.\n\ntigzig.com is now AI-agent First: https://www.tigzig.com/post/tigzig-ai-agent-first-site\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: MCP servers and agents, Browse all apps, Quant apps hub\n\nMore posts: TigZig is Now AI-Agent First, The Fastest Way to Use Tigzig.com? Don't Click. Point Your Claude / AI Agent at It., AI Is the Front Door Now. Don't Dig Through Tigzig.com - Ask Your AI About It., New MCP Hub on Tigzig.com - 7 MCP Servers, 6 Live Apps, 16 Guides, 108-Point Security Checklist. Just Ask Your AI."
},
{
"slug": "tigzig-ai-agent-first",
"title": "TigZig is Now AI-Agent First",
"tags": [
"ai-coders",
"infrastructure"
],
"content": "TigZig is Now AI-Agent First\n\nTigZig AI Agent First\n\ntigzig.com is now AI-agent first. AI coders and agents are first class citizens here. Humans welcome too.\n\nHow does this help you? Instead of clicking around looking for an app or source code or a guide - just ask your AI coder. It can pull the right app link, explain how it works in plain words, and if you want - implement, customize and deploy it for you.\n\nIt has access to all source codes, all 40+ apps & 155+ guides, everything indexed and structured.\n\nBest way to try.. ask your AI Coder (Claude Code, Cursor) to go to tigzig.com and ask:\n'Scan this site. Anything I can productize or sell as a service?'\n'I heard there's a way to connect Excel to databases via MCP. Explain that to me and then set it up also.'\n'I need a full stack app that connects to DuckDB, handles 100s of millions of rows. Find the source code and deploy it.'\n\nYour AI coder will figure out the rest.\n\n40+ live tools. Open source. All indexed for agents.\ntigzig.com\n\nBuilt using llms.txt standard and a bunch of text indexes. Whole site content - 20,000+ lines (not counting codes) - downloadable as a single text file.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, BRIQ\n\nExplore: Browse all apps, MCP servers and agents, Database AI apps\n\nMore posts: tigzig.com is AI-agent first. But what happens when your AI coder runs into a problem on my site?, The Fastest Way to Use Tigzig.com? Don't Click. Point Your Claude / AI Agent at It., AI Is the Front Door Now. Don't Dig Through Tigzig.com - Ask Your AI About It., New MCP Hub on Tigzig.com - 7 MCP Servers, 6 Live Apps, 16 Guides, 108-Point Security Checklist. Just Ask Your AI."
},
{
"slug": "tigzig-api-error-hints-read-the-body-sep2026",
"title": "Tigzig APIs Tell You What Went Wrong. If Your Code Reads Only the Status Code, You Never See It.",
"tags": [
"database-ai",
"ai-coders"
],
"content": "Tigzig APIs Tell You What Went Wrong. If Your Code Reads Only the Status Code, You Never See It.\n\nTigzig APIs try to tell you what went wrong when you hit an error.. with customized error hints. However, if your code reads only the status code, you never see any of it. The hint is sitting there in the response body and one line of printing gets you it.\n\nIf you are debugging by hand, just print the whole body. If it is automated code, print out the error block or the specific 'hint' piece. That's it.\n\nNo, not all errors carry hints - I try and cover the most common / recurring errors that I see in the logs ..\n\nAPI catalog: tigzig.com/apis\n\nQuick tips\nDebugging by hand? Just print the whole body. Everything is in there.\nIn code, pull out the error block on purpose. In Python that is r.json()[\"error\"]. In JavaScript, await res.json() and then j.error. That gives you the whole error block.\nWhere the tailored line sits. On the MF API it is in error.hint. On Vigil, Tremor and the database API it is in error.message. That inconsistency is mine and I am sorting it out; taking the block works on all of them today.\n\nHere is why automated code might miss it\nrequests, httpx and aiohttp all return normally on a 400, so nothing stops your code and nothing tells you to look.\nurllib does raise, but str(e) is only \"HTTP Error 400: Bad Request\" and the body is on e.read().\nraiseforstatus() turns a status into an exception, and that exception carries the summary line only, so log the body itself.\nFor curl, reach for --fail-with-body.\n\nQuick reference: the API debugging guides\n\nHere are quick reference links to the main Tigzig API debugging guides. Just copy text from the guide and hand it over to your AI agent to debug, along with the API doc link at the end.\nDemystifying Tigzig API errors. Twenty common failure modes across the mutual fund, macro and red-flag APIs, every one of them taken from real server logs.\nWhy a SQL error is usually the URL it travelled in. The plus sign, line breaks getting eaten, and clients that stop encoding at the first bracket.\nAn ISIN checker for when a fund lookup fails. It separates a malformed identifier from a valid one whose data is simply absent, and where a character has been dropped it often suggests the intended ISIN.\nGetting a 429. Almost never your total, usually many unbounded calls firing at the same instant.\nIs it you or is it me. A status page with 90 days of uptime history per service.\n\nAPI catalog\nFor you: tigzig.com/apis\nFor your agent: api.tigzig.com\n\nTigzig API error hints: the hint is in the response body, not the status code"
},
{
"slug": "tigzig-api-errors-practical-guide-jul2026",
"title": "Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users",
"tags": [
"infrastructure",
"mcp"
],
"content": "Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users\n\nQuick Resources & Documentation Links:\nFor Humans: Read my complete setup guides at the TigZig API & MCP Hub or visit my dedicated Indian Mutual Funds Pillar Hub.\nFor AI Agents: Point your agents directly to my machine-readable API Catalog (RFC 9727) or the central service index at api.tigzig.com.\n\nI run open, keyless APIs for financial data because I want to make integration as fast and painless as possible. If you are calling these APIs, you don't even have to write the queries yourself. This site is built to be completely AI-agent friendly. You can simply copy this entire page (using the copy button in my blog layout) and hand it over to your AI agent (like Gemini or Claude). They can write the code, execute the calls, and troubleshoot any errors for you.\n\nIf you are using MFPRO - my flagship mutual fund API - or combining it with TREMOR (for live market tracking) and VIGIL (for India corporate red flags and alerts), you are working with an incredibly powerful dataset. But a powerful dataset is only as good as the calls you make to retrieve it.\n\nI constantly monitor my server logs to look for ways to improve the API surface. Analyzing these patterns has helped me squash backend bugs, improve performance, and add auto-correction rules (such as self-healing parameter encoding on my mutual fund endpoint). However, the logs also show a large volume of common user-side slips that developers repeatedly encounter, alongside common misunderstandings of what an \"empty\" response actually means (where the API is working perfectly, but the data is naturally blank).\n\nThis guide is designed to explain the mechanics behind these behaviors in plain English, so you can build more resilient code and stop wasting time debugging.\n\nA link that picked up extra characters on the way\n\nThe single most common failure in my logs is punctuation or extra characters tagging along at the end of a URL.\nPunctuation Stuck to URLs (The Copy-Paste Trap)\n\nCopy a link out of a sentence in an email, blog, or chat, and the trailing period or comma often comes with it. If the server sees a dot at the end, it looks for an endpoint named exactly that, fails, and returns a 404. Look at the very end of your link in your code and delete anything that does not belong.\n\n``\nIncorrect URL with trailing period:\nhttps://api.tigzig.com/v1/openapi.json.\n\nCorrect URL:\nhttps://api.tigzig.com/v1/openapi.json\n`\nThe Stray HTML Entity (&)\n\nIf you copy a URL from a web page's source code instead of the raw text underneath, an ampersand might be represented as &. The amp; tags along as a junk parameter name and breaks the call. Ensure your code is using raw, clean & symbols.\n\nThe characters that join parameters got encoded too\n\nA URL uses & and = to join parameters together. If your code library encodes these characters, the joiners break.\nGlued Parameters and Encoding Errors\n\nIf your code library over-encodes characters, & becomes %26 and = becomes %3D. The server sees one long, meaningless parameter and rejects it. While I recently built an auto-correction salvage for this on the MF NAV endpoint (which now quietly works), other endpoints like TREMOR, VIGIL, and Yahoo Finance will fail. Make sure your client encodes only the parameter values, never the joining symbols themselves.\n\n`\nIncorrect (over-encoded parameters on TREMOR):\nhttps://api.tigzig.com/v1/series?ids=ustreasury10y_yield%26from%3D2024-01-01\n\nCorrect:\nhttps://api.tigzig.com/v1/series?ids=ustreasury10y_yield&from=2024-01-01\n`\n\nSomething the call needs is missing, or literally says \"undefined\"\n\nSometimes the URL structure is fine, but the variables inside are incomplete or uninitialized.\nMissing Crucial Inputs\n\nEndpoints like comparisons or historical charts need a date window to work. If you query /compare?symbols=AAPL without start and end dates, the API has no mathematical range to calculate. It returns a 422 error telling you exactly which fields are missing. Always ensure you provide all required parameters.\nThe JavaScript \"Undefined\" Trap\n\nIf your frontend or backend builds a URL using a variable that hasn't loaded yet, JavaScript will write the literal text \"undefined\" or \"null\" into the URL string. The API receives this text as your fund or ticker identifier and fails. If you see these words in your failed URL logs, the bug is always in your calling application - check that your variables are fully loaded before firing the API request.\n\nRight idea, wrong shape\n\nA handful of failures come from a request that is close, but shaped wrong for the target endpoint.\nGET vs. POST Endpoints\n\nEndpoints that accept input files or analyze text are configured as POST-only. If you paste a POST URL directly into a browser address bar, the browser will attempt a GET request and get a \"Method Not Allowed\" error. Use programmatic tools or clients like curl when sending content payloads.\nFolder-Style Path Guessing\n\nDevelopers sometimes guess at path combinations that are not supported. While my API endpoints are flexible, putting variables or resource categories in place of static path configurations will result in a 400 Bad Request error. Keep query variables inside parameters after the ? symbol.\n\n`\nIncorrect (invalid folder path guess):\nhttps://api.tigzig.com/mf/v1/nav/scheme?category=Equity\n\nCorrect:\nhttps://api.tigzig.com/mf/v1/nav?scheme=147704\n`\nComma-Separated Inputs on a Single-Item Parameter\n\nSome endpoints accept a comma-separated list of identifiers; others are built to fetch exactly one asset at a time. If you pass a comma list to a parameter that only expects one, the server reads the whole thing as a single invalid identifier and returns a 400. Whether a parameter takes one value or many is on each API's own docs page.\n\nOn the MF NAV API this now just works either way. scheme= and schemes= (and isin= / isins=) all accept one identifier or several, comma-separated, up to 50 in a call:\n\n`\nBoth work on MF NAV - one identifier or a comma list, up to 50:\nhttps://api.tigzig.com/mf/v1/nav?scheme=127042,119723\nhttps://api.tigzig.com/mf/v1/nav?schemes=127042,119723\n`\n\nA single-identifier call returns that scheme's rows directly; a multi-identifier call wraps the results under schemes with a count, and any identifier that is unknown or malformed comes back in a not_found list rather than failing the whole call. Only a list where every entry is invalid returns a 400.\n\nPlease check each API's own docs page for whether a parameter takes one value or a list, and what the batch limit is. Firing hundreds of single calls is slower for you and it is the fastest way to hit the rate limit in item 21.\nAsking for the Wrong Data Format\n\nI offer files in high-performance formats like Parquet, but only for my bulk, daily database downloads. If you try to append format=parquet to a live, real-time query endpoint, the API will fail because it does not support streaming that format dynamically. Stick to JSON for real-time queries.\n\nMachine addresses are not web pages\n\nMore of what I run now is MCP, meant for AI agents rather than browsers, and that brings its own two traps.\nOpening MCP Links in the Browser\n\nIf you paste my /mcp server endpoint directly into your browser, you will get a 406 Not Acceptable error. MCP uses Streamable HTTP, which requires an Accept: text/event-stream header that browsers do not send. Do not open these URLs in browsers - point your MCP client configuration (in Claude, Cursor, or other agents) directly at the URL. Details on transport configuration are available on the API & MCP Hub.\nUsing Deprecated Paths\n\nI have unified my MCP services under a single canonical path: /mcp. If you are still using legacy SSE paths like /sse or /v1/mcp/sse, you will receive a 405 Method Not Allowed. Update your AI agent configuration files to use /mcp. AI agents can scan the server configuration endpoints listed in the MCP Server Directory to inspect the active URLs.\n\nBorrowing one app's habits on another\n\nMy different apps sometimes use slightly different parameter names for similar ideas, which causes confusion.\nParameter Naming Differences\n\nA URL that works perfectly on one can fail on another purely because the name is different. I have watched someone try start= and end=, a common convention elsewhere, get rejected because this particular API uses since= and to=, try a third convention as a \"fix\", and fail again. Open that specific API's own docs page, copying the exact parameter names from there rather than reusing the shape remembered from a different tool.\n\nAn identifier that looks right but is not\n\nThis shows up most on MFPRO, where the data requirements from the AMFI backend are specific.\nThe ISIN Mismatch\n\nMFPRO fully supports ISIN codes. However, every fund has multiple variants (Direct vs. Regular plans, Growth vs. Income Distribution cum Capital Withdrawal options). Each variant has its own distinct ISIN. The ISIN listed on your broker statement might be a specific transaction variant that AMFI's official NAV database does not track. If you get a \"not found\" error on a valid ISIN, it means the code is real, but it is not the variant AMFI publishes. Use my search endpoint to search for the fund by name; this will give you the AMFI-supported ISIN or scheme code. You can learn more about how I reconcile fund parameters in the official Mutual Fund NAV API Guide or explore metrics interactively on the MF Pro Docs.\nModel-Hallucinated Codes\n\nWhen developers let AI assistants write their code, the models often hallucinate scheme codes (like 99999). They look correct, but they do not exist. The API returns a structured 404 error with a clear message:\n\n`\n{\n \"detail\": \"No NAV data found for identifier 99999.\",\n \"error\": {\n \"code\": \"NOT_FOUND\",\n \"message\": \"No NAV data found for identifier 99999.\",\n \"docs_url\": \"https://api.tigzig.com/mf/v1/docs\",\n \"hint\": \"No scheme with identifier 99999 exists. Look up the code by fund name: https://api.tigzig.com/mf/v1/search?q=![\"TIGZIG](\"/images/blog/tigzigapierrors_onepager.png\")
\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, TREMOR - Macro Stress Signals, VIGIL - India Red Flag Events\n\nExplore: API + MCP hub, Mutual fund analytics hub, MCP servers for AI agents\n\nMore posts: Data Dictionary & Docs Live on MFPRO. Every Field Explained, 21-Metric Methodology, and How the India MF Universe Actually Looks., Tigzig APIs - Reading the Logs, Shipping the Fixes. Status Page Live, R2 Downloads, Self-Correcting 4xxs., One Line of User Feedback, Fixed Within a Day. MFPRO Search Upgraded + Complete AMFI NAV History, 37M Rows, 2006 Onward., Free Mutual Fund NAV API for India - Revamped. AI Ready. 18,000+ Schemes Since 2008 from AMFI., My Public MCP Server Got Hammered - Security Lessons from a Bot Attack"
},
{
"slug": "tigzig-api-rate-limits-raised-aug2026",
"title": "TigZig API Rate Limits Raised Substantially, Across the Edge, the Proxy and the Apps. And a 429 Now Tells You What to Do About It.",
"tags": [
"infrastructure",
"mutual-funds"
],
"content": "TigZig API Rate Limits Raised Substantially, Across the Edge, the Proxy and the Apps. And a 429 Now Tells You What to Do About It.\n\nSome of you would have run into 429s, the rate limit error, on the TigZig APIs over the last few weeks...especially with Excel-Google sheet trackers and automated scripts. The limits have now been raised substantially, right through the path, the Cloudflare edge, the proxy that sits in front of the apps, and the apps themselves, with the concurrency limits raised alongside.\n\nThese limits were set a long time back when the traffic was a small fraction of what it is now, and they had accumulated across several layers that each went in at a different time. Traffic has risen sharply over the last few weeks and those old numbers started getting in the way of perfectly ordinary use.\n\nThe other thing that has changed is that a 429 now tells you what to do about it, so you get Retry-After and the rate limit headers back and your script or your agent can back off sensibly. And wherever a bulk download is available, the response body names the bulk download that would replace whatever it was you had been looping over.\n\nThe per-API limits are all in the docs / catalogs.\nfor humans: tigzig.com -> API / MCP\nfor agents: api.tigzig.com\n\nand I am watching the error rows while this settles down, and will tweak the dials wherever I can.\n\ntigzig.com/apis\n\nSome quick points for automated callers, scripts and trackers\nOn the 429 itself, every refusal now carries Retry-After, and CORS headers go out on it so a browser fetch can read the status instead of throwing a generic network error at you. Where the refusal comes from the Cloudflare edge you also get the full set, X-RateLimit-Limit, Remaining, Reset and Domain, with Reset being approximate since the window rolls per edge location. An app-side refusal carries Retry-After without that set, and on the mutual fund API that is the one you are more likely to meet.\nIf you are pulling any of this into a spreadsheet based tracker .. say a formula on each row, so a 75 row sheet fires 75 calls on a refresh. At the old limits that was going to be refused part way through. At the new ones it fits with room to spare.\nAlso keep in mind that the APIs offer batch options and for MF, VIGIL & TREMOR you also have the option of downloading the full database ... via single API call or manually from the website. MF link below: mfpro.tigzig.com/dashboard#returns/data-api/download\nPlus, for Excel a simple way to automate is with Python in Excel with xlwings Lite. Guides, tools and downloaders available on tigzig.com -> xlwings Lite.\n\nWhat per IP means in practice\n\nA note on what per IP means in practice, since some of you will be sitting behind a shared address.\n\nThe counting is done against the public IP a request arrives from, which is fine for most callers, but if you are on a mobile network doing carrier-grade NAT, or behind a corporate gateway, then you are sharing that address with other people and in principle their calls are counted against the same budget as yours.\n\nI have not seen that happen in the logs so far and the higher limits make it a good deal less likely, though I am aware it can happen, and if it starts showing up in the error rows I will work out what to do about it then.\n\nThe other side of the same coin is that with no key and no auth on any of this, nothing stops a caller rotating addresses to get around a limit, which is a deliberate choice on my part, because these are public datasets and I want them reachable from a script, a spreadsheet or an agent without anyone signing up for anything, so I would rather ask people to be reasonable about it than put a login in front of open data.\n\nIf you are hitting a limit often enough to be thinking about ways around it, the batch APIs and single file is what you are looking for .. both options available on tigzig.com APIs.\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: API + MCP hub, MF NAV API docs, Mutual fund analytics hub\n\nMore posts: Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users, Your Agents and Scripts Have Been Talking to Me. Here Is That Conversation in Plain English., MFPRO Scheme Master: Every India Mutual Fund Scheme in One Small File, Free, MFPRO Mutual Fund API Update. Three Things Your Calls Asked For, Now Built., 1 in 5 Bulk Downloads Were Silently Truncating. Two Cloudflare Worker Bugs, Found and Fixed. Now Up to 4x Faster."
},
{
"slug": "tigzig-api-status-page-aug2026",
"title": "Getting Errors on the Tigzig APIs? Is It You, or Is It Me? Now You Can Just Go and Look.",
"tags": [
"infrastructure",
"database-ai"
],
"content": "Getting Errors on the Tigzig APIs? Is It You, or Is It Me? Now You Can Just Go and Look.\n\nGetting errors on the Tigzig APIs? MF NAV, Yahoo Finance, Cricket Database? Is it you, or is it me? Now you can just go and look: status.tigzig.com\nAll major APIs covered\n90 days of history\nWhen something breaks my phone buzzes, within a minute or two\nRed on that page usually means I am already on it\nA note at the top with my update\n\nStill getting errors? Just point your AI agent at the OpenAPI docs, they will figure it out. And if something is still stuck, point them at my guides with the most common error patterns and best practices, linked below.\n\nThe page is served from object storage through a Cloudflare Worker, and nothing in that path touches the server it is watching.\n\nSome useful resources if you are building against these\n\nAPI Hub. For you: tigzig.com/apis. For your agent: api.tigzig.com, which gets the full machine-readable catalog plus the spec, the formats and the limits.\n\nBulk downloads, if you would rather have the whole dataset than call an endpoint. MF NAV, Vigil, Tremor and the full cricket database are all downloadable, by API as well as manually. See the docs hub above.\n\nThe cricket database is the newest one, with ball by ball data for ODI, T20 and Test since 2001, 6,900+ matches: tigzig.com/post/cricket-downloads-plain-path-urls-aug2026\n\nCommon errors and best practices\n\nTake a look if you are getting stuck, or just share these with your AI and have it do the work.\n\nSame data in fewer calls. tigzig.com/post/mfpro-nav-fewer-calls-batch-files-parquet-aug2026\n\nDemystifying TIGZIG API Errors. tigzig.com/post/tigzig-api-errors-practical-guide-jul2026\n\nMF NAV API FAQ. tigzig.com/post/mf-nav-api-faq-aug2026\n\nThe Tigzig API status page\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, TREMOR - Macro Stress Signals, DATS-4 Database AI Suite\n\nExplore: API and MCP catalog, Security checklist, Claude for Analytics hub\n\nMore posts: Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users, Getting a 429 From the MF NAV API? It Is Almost Never the Total. Two Changes Fix Nearly Every Case., MF NAV API FAQ. Why Am I Getting an Error, Why Is There No SIF NAV, and Why Does the AUM Look Wrong?, The Cricket Download Files Now Sit on a Plain Path. Append the File Name and That Is the Whole URL., Pulling Indian Mutual Fund NAVs One Scheme at a Time? Three Ways to Get the Same Data in Far Fewer Calls."
},
{
"slug": "tigzig-api-zero-rate-limited-aug2026",
"title": "Zero Callers Rate Limited in 24 Hours. 6,500+ Calls and 9.7 GB Served, After the TigZig API Limits Went Up.",
"tags": [
"infrastructure",
"mutual-funds"
],
"content": "Zero Callers Rate Limited in 24 Hours. 6,500+ Calls and 9.7 GB Served, After the TigZig API Limits Went Up.\n\nZero API callers rate limited on the TigZig public APIs in the last 24 hours. 6,500+ calls, 9.7 GB served, zero refusals, after the limits went up.\n\nThis is after a short period where at the peak, one in four API calls was getting refused. Old rate limit and concurrency configurations never got updated. Now fixed across the pipeline from the edge, proxy to the app.\n\nAnd the API servers have been busy busy lately...after chilling out for a long time when there were very few users.\n\nAnd the three busiest surfaces get used in completely different ways.\n\nMutual funds (MFPRO) is the largest by call volumes and most callers use it two ways. Some callers pull NAVs scheme by scheme (or batches), others take the whole database in a single file, along with the scheme master - which by the way is also updated 3x a day with latest AMFI data. Roughly equally divided between parquet and CSV format. Most of the calls to the NAV path look like scripts on a schedule, most likely to update trackers and tools. And that's exactly one of the uses the API was designed for.\n\nYahoo Finance (YFIN) is the 2nd largest by call volume, and get-all-prices is the most used path. By the way, it's a thin wrapper over the yfinance package, and the code is open if you want to run your own.\n\nTremor is almost entirely bulk. Callers take the whole macro database, global and India, a few hundred indicators.\n\nDownloads carry their own limit, tighter than the query limits, because a download is heavy. Range requests have started arriving, so that limit is in the process of being revised significantly upwards as we speak.\nAPI Docs (humans): tigzig.com/apis\nAPI Docs (agents): api.tigzig.com\nMFPRO Live Tool: tigzig.com/mfpro\nQuants Report Agent: quants.tigzig.com\n\nHand it to your agent\n\nIf you would rather not click through any of this, hand it to your agent.\n\nPaste into Claude, ChatGPT or any other AI Agent (Cowork / Coders):\n\n``\nGo to api.tigzig.com. I want to pull prices from yahoo finance and navs for\nmutual funds and update my tracker. help me set it up end to end. but first\ngo and test the apis that they work\n``\n\nPrefer a screen? The front ends are free too\n\nEverything above is the API side. If you would rather use a screen, the front ends are free too, and no login on any of them.\nMFPRO. Mutual funds, the biggest one. Compare up to 200 funds at a time on returns and risk, rolling returns, SIP and category benchmarking, and each of the 21 metrics has its own page with the formula, a worked example and how to read it, computed on AMFI NAV data. tigzig.com/mfpro\nQuants Agent. Portfolio stats, an AI technical chart read, and Yahoo Finance extraction for any ticker, out as a PDF or a web report. tigzig.com/n8n-tech-analysis\nQREP. Compare securities against a benchmark on 81 QuantStats metrics, CAGR, Sharpe, Sortino, max drawdown and the rest. tigzig.com/qrep\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, QRep - Security Reports, Quants Agent\n\nExplore: API + MCP hub, MF NAV API docs, Mutual fund analytics hub\n\nMore posts: TigZig API Rate Limits Raised Substantially, Across the Edge, the Proxy and the Apps. And a 429 Now Tells You What to Do About It., Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users, Your Agents and Scripts Have Been Talking to Me. Here Is That Conversation in Plain English., MFPRO Scheme Master: Every India Mutual Fund Scheme in One Small File, Free, Claude Cowork Runs a Real Python Sandbox. I Put 37 Million Rows of Mutual Fund NAV Data Through It."
},
{
"slug": "tigzig-apis-data-open-jun2026",
"title": "The Data Behind My Analyses Is Now Open. Download the Whole Database or Query a Single Row.",
"tags": [
"portfolio-quants",
"mcp"
],
"content": "The Data Behind My Analyses Is Now Open. Download the Whole Database or Query a Single Row.\n\nPublished: 2026-06-05\n\nThe analyses I've been posting these past months all run on the same data - and that data is now open to everyone. US delinquencies, private credit, life insurance, the markets piece - all off primary data, FRED, RBI, Fed, FDIC, refreshed three times a day by the same pipeline that runs the live apps. Three open APIs powering the Tigzig apps: MF NAV (18,000+ schemes, 13 yrs, ~20M daily NAVs), VIGIL (corporate signals: ratings, insider, pledge, SAST, RPT), TREMOR (~300 macro indicators). Pull a single row across ~30 endpoints, or take the whole database from one endpoint in SQLite, DuckDB, Parquet, CSV or TSV - refreshed 3x a day by the CI/CD that powers the apps. Connect to the live MCP server, or wrap any of these APIs in your own MCP. No auth. Rate limited. Free to use. For humans: tigzig.com -> APIs. For agents: agents.tigzig.com.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/tigzig-apis-data-open-jun2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/tigzig-apis-data-open-jun2026 with header Accept: text/markdown\n\nTags\nportfolio-quants, mcp\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, VIGIL - India Red Flag Events, MFPRO - Mutual Fund Analytics\n\nExplore: APIs and MCP hub, Database AI apps, Market tools\n\nMore posts: VIGIL Data API and Download - 315K+ Records, 7 Tables, Updated Daily, Free Mutual Fund NAV API for India (AMFI) - No Key, Download 20.5M+ NAVs in One Call, Tigzig Tremor v2 Is Live. 330+ Macro + Credit Indicators Behind an MCP. 83% Lighter Startup. Just Ask Your AI Agent., Q1 2026 US Lending Data Live on Tigzig. NCUA + FDIC + NY Fed in One API. Two of Three Not on FRED. AI-Agent Ready., New MCP Hub on Tigzig.com - 7 MCP Servers, 6 Live Apps, 16 Guides, 108-Point Security Checklist. Just Ask Your AI."
},
{
"slug": "tigzig-apis-status-page-live-jul2026",
"title": "Tigzig APIs - Reading the Logs, Shipping the Fixes. Status Page Live, R2 Downloads, Self-Correcting 4xxs.",
"tags": [
"mutual-funds",
"infrastructure"
],
"content": "Tigzig APIs - Reading the Logs, Shipping the Fixes. Status Page Live, R2 Downloads, Self-Correcting 4xxs.\n\nPublished: 2026-07-08\n\nRelease note on what shipped for the Tigzig public APIs in the last 48 hours, driven by daily log reading (200s tell what is working, 4xx and 5xx tell what to fix next). Four things live: (1) common-guess coercion - if a caller or their AI guessed /series instead of /nav or sent start= instead of since=, those now quietly work with one canonical spec still in the docs; (2) every 4xx now returns a help block linking the docs page and the machine catalog so an AI agent can self-correct without the human stepping in; (3) full-database downloads (100 to 500+ MB files) now served from Cloudflare R2, noticeably faster than before; (4) status page live and probed every minute at status.tigzig.com - downtime rings the author's phone. Underway: fallback to a secondary server if the primary goes down. Second half of the post is a practical errors-and-fixes list: 404 (scheme code guessing/enumeration - use /search?q=fund-name or an ISIN from the broker statement, both work), heavy parallel bursts now return proper 429s instead of the occasional 503, 429 rate limits (300/min per public IP with X-RateLimit headers so scripts/agents can pace themselves), 400 validation (YYYY-MM-DD dates, 50-scheme bulk cap, scheme= for one vs schemes= for many), and every failure comes back as JSON with an error code + docs_url pointing straight at the fix. Polling tip: send If-Modified-Since and you get a 304 on days AMFI hasn't published (weekends, holidays) so scripts can skip re-downloading history they already have. Single-image release post - status page screenshot.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/tigzig-apis-status-page-live-jul2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/tigzig-apis-status-page-live-jul2026 with header Accept: text/markdown\n\nTags\nmutual-funds, infrastructure\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, Quants Agent\n\nExplore: Full API + MCP catalog, MF NAV API docs, Mutual fund analytics hub\n\nMore posts: Free Mutual Fund NAV API for India - Revamped. AI Ready. 18,000+ Schemes Since 2008 from AMFI., Copy Page as Markdown - Every Page on Tigzig.com Now Hands Your AI Full Context in One Click., Free AMFI Mutual Fund NAV API for India - Historical NAV by Scheme Code, No Key"
},
{
"slug": "tigzig-ask-chatgpt-stop-browsing-jun2026",
"title": "Stop Browsing Tigzig.com. Ask ChatGPT First. (Or Whichever AI You Use.)",
"tags": [
"ai-coders"
],
"content": "Stop Browsing Tigzig.com. Ask ChatGPT First. (Or Whichever AI You Use.)\n\nPublished: 2026-06-26\n\nStandalone one-pager release: tigzig.com now has 300+ assets (live apps, repos, build guides, data APIs, MCP servers, analysis) and is indexed across ChatGPT, Claude, Gemini and Meta AI on WhatsApp. Don't click through it the old way - just ask whichever AI you keep open, and it walks you through what is there. Sample agent prompts in the deck: 'tigzig.com - anything on Python in Excel?', 'tigzig.com - anything on Claude in Excel?', 'tigzig.com - any guides on security for AI tools?', 'tigzig.com - any portfolio analysis tools I can use?', 'tigzig.com - pull the Claude Cowork guide', 'tigzig.com - what is on mutual fund analytics?', 'tigzig.com - any infra guides for tool builders?'. Faster than clicking around, and a fair way to sanity-check the site too. All free, AI-agent-first.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/tigzig-ask-chatgpt-stop-browsing-jun2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/tigzig-ask-chatgpt-stop-browsing-jun2026 with header Accept: text/markdown\n\nTags\nai-coders\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, Practice Lab\n\nExplore: Browse all apps, API and MCP hub, Guides\n\nMore posts: AI Is the Front Door Now. Don't Dig Through Tigzig.com - Ask Your AI About It., The Fastest Way to Use Tigzig.com? Don't Click. Point Your Claude / AI Agent at It., New MCP Hub on Tigzig.com - 7 MCP Servers, 6 Live Apps, 16 Guides, 108-Point Security Checklist. Just Ask Your AI., Claude Cowork and Claude Code Have Quietly Become My Universal Front End. More Than Excel, Jupyter or Colab."
},
{
"slug": "tigzig-ask-your-ai-front-door-jun2026",
"title": "AI Is the Front Door Now. Don't Dig Through Tigzig.com - Ask Your AI About It.",
"tags": [
"ai-coders",
"mcp"
],
"content": "AI Is the Front Door Now. Don't Dig Through Tigzig.com - Ask Your AI About It.\n\nPublished: 2026-06-18\n\nStandalone one-pager release: AI is the front door now. Just like you ask your AI about Cloudflare, Vercel or AWS instead of digging through the docs, tigzig.com (40+ live apps, repos, ~200 build guides, data APIs, MCP servers, 100s of assets) is indexed across the major AI assistants - ChatGPT, Gemini, Google AI, Claude, WhatsApp Meta AI - so you can ask whichever one you keep open what's there, what's not, and how to use it before you even visit. Sample questions: Is tigzig.com legit? What can it do with my database? Does it have a security checklist for AI tools? How do I pull a portfolio or technical-analysis report? What free data APIs does it have? Quicker than browsing, and a fair way to sanity-check it.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/tigzig-ask-your-ai-front-door-jun2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/tigzig-ask-your-ai-front-door-jun2026 with header Accept: text/markdown\n\nTags\nai-coders, mcp\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, BRIQ\n\nExplore: Browse all apps, APIs and MCP servers, Security checklist\n\nMore posts: Stop Browsing Tigzig.com. Ask ChatGPT First. (Or Whichever AI You Use.), The Fastest Way to Use Tigzig.com? Don't Click. Point Your Claude / AI Agent at It., tigzig.com is AI-agent first. But what happens when your AI coder runs into a problem on my site?, New MCP Hub on Tigzig.com - 7 MCP Servers, 6 Live Apps, 16 Guides, 108-Point Security Checklist. Just Ask Your AI."
},
{
"slug": "tigzig-claude-for-analytics-hub-jul2026",
"title": "Claude for Analytics on Tigzig.com - New Hub: Excel, PowerPoint, Cowork, 7 MCP Servers, 20+ Guides.",
"tags": [
"claude-in-excel",
"mcp",
"excel"
],
"content": "Claude for Analytics on Tigzig.com - New Hub: Excel, PowerPoint, Cowork, 7 MCP Servers, 20+ Guides.\n\nNew hub on tigzig.com: Claude for Analytics - in Excel, PowerPoint and Cowork.\n\nEverything I have built and written on driving analyst work with Claude, pulled onto a hub page. 20+ guides, 5 PDF playbooks, 7 MCP servers. Plus the workbooks.. formulas, charts, models. Open them and poke around.\n\nTalk to Postgres and DuckDB from inside Excel. Pull 81 risk and return metrics live into a sheet. Work with Claude-in-Excel and have it push results to Claude-in-PowerPoint. And Cowork takes the same Claude beyond the sheet.. files, databases, ML models, code, one frontend for all of it.\n\nBrowse it: tigzig.com -> Claude for Analytics.\n\nOr just ask your Claude, ChatGPT or Gemini to go read the page and tell you what's there.\n\nAll free. Almost all open source.\n\nSome practical notes on which Claude I use for what\n\nMy primary tool is still Claude Code. More than half my work now is full stack tool builds.. that is all Claude Code only. Same for the heavy data work, 100s of PDFs, scraping, deployments, server, security.\n\nBut a lot of the day to day has moved. Excel-heavy work goes directly to Claude in Excel. Claude Code / Cowork can work on Excel files, but that is not the effective route. Decks and reports, unless it must be a PPT, I now do in Cowork as HTML and transport to PDF. No PowerPoint in between. Claude in PowerPoint does the job but it plays around with objects and formats.. takes time.\n\nML models are mostly Claude Code, some in Cowork. I have hardly touched a Jupyter notebook in months. I tell it exactly what I want, from data cleaning and initial diagnostics through final test validations, and ask for all results back as HTML. All my discussion with Claude happens over HTML now. Very fast.\n\nAnd do read Anthropic's own guides.. their official documentation is very nicely written, simple plain language. Links are in the official docs section on the hub page.\n\nEarlier I used to get confused which one to use where. Now it is instinctive. Try it. Use it. Give it time.\n\nOn AI supervision, and on paying for this stuff\n\nAI will do all the heavy lifting.. analysis, ML models, dashboards. But you have to catch its mistakes, in the logic or on the business side. Left unguided it takes its own calls.. an ML model will overfit, hand you accuracy and confusion metrics when you wanted a decile table, run nestimators to a thousand when 100-150 trees with maxdepth of two was what you wanted. You have to tell it. Inspect what you expect. You need to know your stuff.\n\nPricing, my honest opinion. $20 plan is fine for starting off and light work. Decent regular work, you will need the $100 Max. Heavy end to end work.. tool builds, full stack apps.. nothing below the $200 Max works. At $200 you get basically a whole team of data scientists and developers combined.\n\nBut not everybody needs the $200. Among my clients, a few key folks are on it, others do solid work on Cursor at $20 (composer is blazing fast). Occasional use, free limits might do.. Codex is coming up well, Antigravity and Copilot too. Match the tool to the work. High stake work - pay up.\n\nFor enterprises, it's API based - probably costs ~50x the subscription price per employee. Same concept, different scale.\n\nAmar Harolikar · Decision Sciences & Applied AI · tigzig.com.\n\n\nRelated\n\nTools: Practice Lab, Databases + ML Models, QRep - Security Reports\n\nExplore: Claude for Analytics hub, How I Work, API / MCP catalog\n\nMore posts: New MCP Hub on Tigzig.com - 7 MCP Servers, 6 Live Apps, 16 Guides, 108-Point Security Checklist. Just Ask Your AI., Claude Cowork and Claude Code Have Quietly Become My Universal Front End. More Than Excel, Jupyter or Colab., QREP in Excel: 81 Risk and Return Metrics Inside Your Sheet Through Claude in Excel., Power User Guide to Claude in Excel & PowerPoint - 26 Working Tips"
},
{
"slug": "tigzig-copy-page-button-jul2026",
"title": "Copy Page as Markdown - Every Page on Tigzig.com Now Hands Your AI Full Context in One Click.",
"tags": [
"ai-coders",
"converters-tools"
],
"content": "Copy Page as Markdown - Every Page on Tigzig.com Now Hands Your AI Full Context in One Click.\n\nYou are reading an API page or an Excel-MCP guide on tigzig.com and you want your AI to just take it from there, explain it, write the code, set the whole thing up on your machine. That is now one click.\n\nEvery page on the site has a Copy page button now, and it hands your AI the entire page as clean markdown, full context in one go.\n\nCopy the MF NAV API page, paste it into Claude Cowork, ask for a SIP analysis, and it reads the page, calls the API and runs the whole thing for you.\n\nAnd if you do not feel like browsing at all, just ask your ChatGPT or Gemini or Claude directly, something like \"go to tigzig.com and find what to do for my Excel to talk to my database.. and help me deploy it also\".\n\nNo hunting through the site, no screenshots, no partial copy paste.\n\nSite has been built as AI Agent First. Humans welcome too.\n\nLive example - one paste, no code, a full SIP analysis\n\nPasted the MF NAV API page (https://www.tigzig.com/apis/mf-nav) into Claude Cowork, asked for a 10 year SIP analysis. It searched the scheme by name, pulled 2,464 daily NAVs, ran 120 monthly installments of Rs 10,000 on Parag Parikh Flexi Cap Direct, and came back with Rs 12 lakh grown to Rs 30.5 lakh, XIRR 17.75%. One paste, no code from my side.\n\nCopy page button on tigzig.com - one-pager overview\n\nAmar Harolikar · Decision Sciences & Applied AI · tigzig.com.\n\n\nRelated\n\nTools: Portfolio Analytics Agent, MFPRO - Mutual Fund Analytics, Quants Agent\n\nExplore: MF NAV API doc, API and MCP catalog, Browse all apps\n\nMore posts: AI Is the Front Door Now. Don't Dig Through Tigzig.com - Ask Your AI About It., Free Mutual Fund NAV API for India - Revamped. AI Ready. 18,000+ Schemes Since 2008 from AMFI., Claude for Analytics on Tigzig.com - New Hub: Excel, PowerPoint, Cowork, 7 MCP Servers, 20+ Guides."
},
{
"slug": "tigzig-excel-to-databases-ai-find-jun2026",
"title": "Connect Excel to a Database with Tens of Millions of Rows? Tigzig.com Has 15+ Apps and Guides for That Alone. Just Ask Your AI.",
"tags": [
"excel",
"python-in-excel",
"mcp"
],
"content": "Connect Excel to a Database with Tens of Millions of Rows? Tigzig.com Has 15+ Apps and Guides for That Alone. Just Ask Your AI.\n\nPublished: 2026-06-19\n\nStandalone one-pager release: how to connect Excel to a database with tens of millions of rows and GBs of data. There are 15+ apps, repos and guides on tigzig.com just for this one niche - xlwings-Lite (Python in Excel), a Claude-in-Excel MCP server, open repos connecting to MySQL, PostgreSQL and DuckDB, plus security setups, auth and semantic layers. And Excel-to-databases is only one corner - the whole site has 250+ assets, from live tools to open repos. The point of the post: don't click around. Tigzig.com is indexed across ChatGPT, Gemini, Claude, Google AI and Meta AI on WhatsApp, so just ask whichever AI you keep open what you want to pull from the site - it will find it and walk you through. If you are on Claude Code, Cursor or any AI coder, it will download and implement it for you. Sample agent prompts in the deck: 'What does tigzig.com have for connecting Excel to a database?' and 'What about security, semantic layer, auth?'\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/tigzig-excel-to-databases-ai-find-jun2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/tigzig-excel-to-databases-ai-find-jun2026 with header Accept: text/markdown\n\nTags\nexcel, python-in-excel, mcp\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: Practice Lab, Database AI Apps, DUCKIT - CSV to DuckDB\n\nExplore: Claude for Analytics hub, MCP servers and agents, Browse all apps\n\nMore posts: Talk to Your Database from Excel - Postgres, DuckDB - via Claude in Excel with MCP, xlwings lite |Connect to Remote Databases, MCP Guide for Claude in Excel - and any MCP client, Talk to Your Database from Excel via Claude & MCP - Part 2"
},
{
"slug": "tigzig-mcp-hub-launch-jun2026",
"title": "New MCP Hub on Tigzig.com - 7 MCP Servers, 6 Live Apps, 16 Guides, 108-Point Security Checklist. Just Ask Your AI.",
"tags": [
"mcp",
"security"
],
"content": "New MCP Hub on Tigzig.com - 7 MCP Servers, 6 Live Apps, 16 Guides, 108-Point Security Checklist. Just Ask Your AI.\n\nPublished: 2026-06-26\n\nStandalone one-pager release announcing the new MCP hub on tigzig.com - 7 open-source and live MCP servers, 6 live apps, 16 guides, plus a 108-point security checklist. Designed for AI and automation builders. Either browse the hub at tigzig.com/mcp, or just ask ChatGPT / Gemini / Meta AI on WhatsApp / Claude something like 'Go to tigzig.com's MCP page and lemme know what's there' - the AI will pull what's there and Claude Code / any AI coder will implement it for you. The hub carries reusable components that work across domains - sending data and charts to AI and getting structured results back, formatted PDF and HTML output, accessing backend databases on DuckDB and Postgres, Python processing for statistical and quant analysis. Live MCP servers and apps cover macro data, a database query server, portfolio analytics, and tools used regularly for own analysis. The companion security checklist (tigzig.com/security) covers backend, auth, database, edge and third-party apps - 108 items.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/tigzig-mcp-hub-launch-jun2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/tigzig-mcp-hub-launch-jun2026 with header Accept: text/markdown\n\nTags\nmcp, security\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QRep - Security Reports, BRIQ\n\nExplore: MCP + API catalog, Security checklist, Database AI apps\n\nMore posts: Claude for Analytics on Tigzig.com - New Hub: Excel, PowerPoint, Cowork, 7 MCP Servers, 20+ Guides., New Markdown to PDF Hub on Tigzig.com - ReportLab, Repos, Live Tools, MCP + API Endpoints., Tigzig Tremor v2 Is Live. 330+ Macro + Credit Indicators Behind an MCP. 83% Lighter Startup. Just Ask Your AI Agent., AI Is the Front Door Now. Don't Dig Through Tigzig.com - Ask Your AI About It."
},
{
"slug": "tigzig-mcp-streamable-http-migration-jul2026",
"title": "Every Tigzig MCP is now Streamable HTTP. Requesting existing users from LM Studio, ChatGPT, Claude etc to make a one time change if you have not already done so",
"tags": [
"mcp"
],
"content": "Every Tigzig MCP is now Streamable HTTP. Requesting existing users from LM Studio, ChatGPT, Claude etc to make a one time change if you have not already done so\n\nI finished moving all my MCP servers to Streamable HTTP, which is the current transport standard that Claude, ChatGPT, LM Studio and other clients use today. It's been running alongside the earlier SSE for the past month or two. I have tried my best to maintain backwards compatibility with old URL's but suggest to make a one time connection change if you are still on the old one.\n\nFor all new users - you should use the Streamable HTTP URL, because that is where all the new tools and improvements would be implemented.\n\nAs I add new tools or improve the backends, you get those updates automatically, and you never have to touch the configuration URL again.\n\nEvery MCP now lives at its host address plus /mcp.\n\nAnd as a quick aside, it is great to see the real logs here, because I can watch agents connecting from Claude, ChatGPT, and LM Studio daily, and it is clear that people are actually using these setups.\n\nBackward Compatibility\n\nI tried to make this migration as clean as possible, and I am keeping backward compatibility on two fronts.\n\nFirst is the transport. Streamable HTTP is the new standard, but the older SSE connection remains mounted and active alongside it, so an existing SSE setup should not break. However keep in mind that the SSE URLs are also in the process of deprecation and would be phased out soon.\n\nSecond is the old addresses. I am keeping them running where possible, so setups that were already working can keep working. I say where possible because standardizing the endpoints did move a few paths. If one of your setups is not connecting, you should just grab the current URL from the list and paste it in, rather than trying to debug the old path. The docs are always the source of truth for the live endpoints.\n\nA Quick Word on Connection Types\n\nWhen you add one of these servers to Claude, ChatGPT, LM Studio etc, the setup screen will ask you to select a connection type before pasting the URL, and that part can be confusing. There are only two details you need to know.\n\nFirst, these are all remote servers. They run on my machines rather than your computer, and you reach them over the internet. There is nothing to install locally, so if your tool asks you to choose between a local command and a remote URL, you want the remote URL option.\n\nSecond, for a remote connection, your app will offer you two styles, which are SSE and Streamable HTTP. You should select Streamable HTTP, or simply HTTP if that is how it is labeled in your client, and then paste the URL from the list. The setup is just those two steps.\n\nWhy use Streamable HTTP instead of SSE? The MCP ecosystem has settled on Streamable HTTP as the transport standard, and the protocol is being evolved rather than replaced. SSE was the earlier style, and while I am keeping it mounted for the time being, the wider community is moving away from it, so there is no reason to use it for new setups.\n\nThe MCP Servers\n\nAll of these servers are live and configured with Streamable HTTP:\nYahoo Finance: https://yfin-h.tigzig.com/mcp\n Stock market data, company financials, historical prices, analyst estimates, and ownership. Covers global and Indian tickers.\nTechnical Analysis: https://ta.tigzig.com/mcp\n Technical analysis indicators for any ticker, with automated AI commentary.\nPortfolio Review: https://ffn.tigzig.com/mcp\n Portfolio performance and risk metrics built on the open-source ffn library.\nCompare Securities (QRep): https://qrep-api.tigzig.com/mcp\n Side-by-side comparison of up to six securities against a benchmark, using QuantStats metrics.\nTigzig Data Surface: https://api.tigzig.com/mcp\n Live macro and market datasets from FRED, RBI, ECB, OECD, and MoSPI.\nCricket Database: https://db-mcp.tigzig.com/mcp\n Read-only SQL queries over ball-by-ball T20 and ODI matches from 2013 to 2025.\nMarkdown to PDF: https://mdtopdf.tigzig.com/mcp\n A utility that converts raw markdown into a formatted PDF document.\n\nOne note: pasting an /mcp address into a browser will show a 406 Not Acceptable error. That is expected, these are machine endpoints for MCP clients, not web pages. Point your MCP client at them, not your browser.\n\nSkipped the MCP?\n\nIf you do not want to connect to a third-party MCP server, you have two simple alternatives:\nEvery MCP is just a wrapper around an open API. You can skip the MCP layer entirely and call the APIs directly using standard HTTP requests.\nBecause the APIs are open, you can easily wrap your own MCP server around them.\n\nAll of these endpoints are completely open and require no API keys or signup. There is a public IP rate limit to prevent abuse, and you can find the specific limits documented on each individual API page. That is the entire setup.\n\nOther Open Services\n\nI also updated and cleaned the documentation across all these endpoints. If you want to point an AI agent or a developer tool at these resources, you can give it the machine-readable directory files and it will map the tools automatically.\n\nFor your AI agents: point them at https://api.tigzig.com. That one address lists every API and every MCP server, with a machine-readable catalog and an agent guide. They will find their way from there.\n\nIf you'd like to browse yourself, then head over to: https://www.tigzig.com/apis\n\nTigzig Data Surface, the same server listed in the MCP section above, is a single public read-only API that exposes ~330 curated macro, credit, valuation, insurance, India, global, markets and FX indicators across 8 categories, sourced from 17+ publishers. No authentication. It is the same data that backs the TREMOR app.\nhttps://api.tigzig.com/mcp\n\nIndia Mutual Fund NAV API. A free, open AMFI mutual fund NAV API for India - no key, no sign-up. AMFI publishes NAVs as a giant daily text dump. The API lets you download the entire NAV history - every scheme, every day - in a single call. Or you can pull one scheme at a time, or hit the batch endpoint to download up to 50 schemes in a batch.\nhttps://api.tigzig.com/mf/v1/\n\nVigil, which is my India corporate red-flags tracker, is also open at https://api.tigzig.com/vigil/v1/\n\nIf your app is still returning an error, especially if you are using the API directly, I wrote a short guide detailing the most common setup mistakes .. it is usually one of those simple issues, and you can read the troubleshooting steps here:\n\nhttps://www.tigzig.com/post/tigzig-api-errors-practical-guide-jul2026\n\nBeyond the MCP servers: just ask your AI\n\nYou do not have to dig through the whole site. Beyond the MCP servers there are 300+ assets on the site, tools, guides and walkthroughs, and it is all built so your own AI can find the right one.\n\nJust ask your ChatGPT, Gemini or Claude. A few prompts to try:\n\"On tigzig.com, which tool compares several securities on price performance and risk against a benchmark, and is it available as an MCP server? Give me the URL.\"\n\"I would rather not use Tigzig's hosted MCP, I want to run my own with Google OAuth in front of it. Does tigzig.com have an open API and any guides I can use to build my own MCP around it?\"\n\"My Tigzig MCP is erroring out in my client. Go to tigzig.com, check the MCP and API docs, and tell me the likely fix.\"\n\"I work in Excel and want to run Python there. What tigzig.com guides cover Python in Excel, and which tool should I start with? And any guides to connect to my database directly from Excel also?\"\n\nOne more angle: AI workers, not just AI chatters\n\nOne more angle, and it goes beyond MCP.\n\nThese tools have quietly turned into AI workers, not just AI chatters. In Claude that is Cowork, in ChatGPT it is the work and data-analysis mode, and the AI coders like Claude Code, Codex and Cursor go further still.\n\nAI Workers can connect to your systems, pull data, run models, and build the actual output, an Excel model, a PDF report, an HTML dashboard.\n\nIf you are new to this, the Cowork style worker is the simpler place to start, and the coders are more powerful once you want full builds.\n\nI wrote up the three tiers and which one to reach for when: https://www.tigzig.com/post/ai-worker-not-ai-chatter-jul2026\n\nA power tip: you do not always need an MCP at all. With just a free FRED API key, your AI worker can reach into all 845,000 FRED economic series directly, no plugin, no install, then chart it and drop it into Excel or a report: https://www.tigzig.com/post/fred-your-ai-pulls-directly-jul2026\n\n![\"Every](\"/images/blog/mcpstreamableonepager.png\")
\n\nCricket data from Cricsheet, published under the Open Data Commons Attribution License 1.0 (ODC-BY).\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QRep - Security Reports, Quants Agent\n\nExplore: API + MCP hub, MCP servers for AI agents, Claude for Analytics hub\n\nMore posts: Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users, New MCP Hub on Tigzig.com - 7 MCP Servers, 6 Live Apps, 16 Guides, 108-Point Security Checklist. Just Ask Your AI., MCP Guide for Claude in Excel - and any MCP client, My Public MCP Server Got Hammered - Security Lessons from a Bot Attack"
},
{
"slug": "tigzig-mf-nav-api-revamp-jul2026",
"title": "Free Mutual Fund NAV API for India - Revamped. AI Ready. 18,000+ Schemes Since 2008 from AMFI.",
"tags": [
"mutual-funds",
"converters-tools",
"mcp"
],
"content": "Free Mutual Fund NAV API for India - Revamped. AI Ready. 18,000+ Schemes Since 2008 from AMFI.\n\nRevamped the free Mutual Fund NAV API for India. Made it easier to work with AI.. and for automations. Usage suddenly went up past few weeks.. so I have added more capacity and made the whole thing more flexible.\n\nDaily NAVs for 18,000+ schemes since 2008, straight from AMFI. Free, no sign up, no API key.\n\nYou should see latencies and errors significantly down.. and you can now give both AMFI Code or ISIN or even mix it up - I am auto-detecting at the backend. Plus many more things.. check the two pager attached.\n\nOne thing I will call out - Claude Cowork as well as Codex can work with the API directly. Point them at the docs page and they take it from there and run an analysis based on your directions. Claude in Excel just needs a thin MCP wrapper on top. Open source wrapper coming soon.\n\nAPI Docs: tigzig.com -> APIs -> MF NAV\n\nIf you'd rather use the user interface then tigzig.com -> MFPRO - run advanced analytics on up to 200 schemes in one shot.. risk, returns (including rolling return options), alpha, betas, drawdowns.\n\nWant the whole database? Two ways\n\nHuman way: go to MFPRO app at https://www.tigzig.com/mfpro -> Data & API. Click the button and the file downloads. Parquet, CSV, TSV, or SQLite.\n\nAI / API way: ask your AI to send out one GET request on /mf/v1/download and all 20.5M+ rows land as a single file.\n\nSizes to keep in mind: Parquet about 109 MB, CSV about 132 MB compressed, SQLite about 462 MB compressed. The /downloads/manifest endpoint gives current sizes and row counts, worth a check before you pull. Parquet loads straight into pandas in one line and you can start processing right away.\n\nAnd if you are running your own analysis\n\nEvery metric in MFPRO ships with a validation pack.\n\nRun a preset selection, run the validations, and you get a zip: the validation report, any variances flagged, and the actual Python code that produced the numbers, ready to run as main.py from your local disk against the NAV file.\n\nThe UI computes on DuckDB SQL and the validation recomputes in pure Python, and the pack shows both so you can see they tally. If you want ready-to-use, tested code for these metrics, that is it. Your Claude Code or Cowork can also take that code and set it up fresh for you.\n\nFacing any issue with the API, DM me, use the feedback button on the site, or just tell your AI to check llms.txt on tigzig.com and send me feedback. It will figure it out.\n\nAmar Harolikar · Decision Sciences & Applied AI · tigzig.com.\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, Quants Agent\n\nExplore: MF NAV API docs, Mutual fund analytics hub, Database AI apps hub\n\nMore posts: Free Mutual Fund NAV API for India (AMFI) - No Key, Download 20.5M+ NAVs in One Call, Now live - MFPRO v2 - Mutual Fund Analytics (India) - now with rolling returns, custom eval periods, multi-period & multi-instrument comparisons, The Data Behind My Analyses Is Now Open. Download the Whole Database or Query a Single Row., The Fastest Way to Use Tigzig.com? Don't Click. Point Your Claude / AI Agent at It."
},
{
"slug": "tigzig-private-credit-hub-jul2026",
"title": "New Hub - Private Credit Stress on Tigzig.com. 19 Analyses, 3 Live Tools, Regulator Reports on One Page.",
"tags": [
"portfolio-quants"
],
"content": "New Hub - Private Credit Stress on Tigzig.com. 19 Analyses, 3 Live Tools, Regulator Reports on One Page.\n\nNew hub on tigzig.com: Private Credit Stress. 19 analyses, 3 live tools, the regulator reports.. all on one page.\n\nSo is private credit cracking? Well, depends on who you ask. I remember 2008. Till almost the very end, the mortgage market was either perfectly stable or falling apart..... depending on who you asked. Then Lehman happened.\n\nWhat the last 30 days added: $14B of investor money sitting behind 5% redemption gates. One Blue Owl fund got requests for 38% of its shares. A BlackRock BDC chief exiting mid federal valuation probe.\n\nAnd PIMCO calling out a \"confidence gap\".. NAVs increasingly reflecting manager marks rather than a shared clearing price.\n\nThe question has moved from \"are borrowers defaulting\" to \"can we trust the marks.\"\n\nBrowse it: tigzig.com -> Private Credit.\n\nOr just ask ChatGPT, Gemini or Claude \"Go to tigzig.com's private credit page and lemme know what's there\".\n\nRaw data behind the tools, downloadable. Source links for the external reports.\n\nThe real story: in just one month supervisors moved from writing warnings to running drills\n\nECB doubled its private credit probe to 20+ banks, with annual reporting for those with meaningful exposure (Jun 15). France ran its first ever system-wide NBFI stress test.. Banque de France, ACPR and AMF, 25 institutions, fire-sale and interconnection risk (Jun 17). Bank of England sent 46 firms a hypothetical five-year recession to model how private markets behave under stress (Jun 19). And at Sintra, Lagarde and Bailey both flagged non-bank oversight in the same week.\n\nWarnings are cheap. Stress tests cost money and staff. When three central banks fund the drills in the same month, they are not idly curious.\n\nTo be fair.. the gates are working as designed so far. No fire sales, and a few funds (Goldman, Oaktree) saw redemption demand below their caps. That is the bull case, and it is not nothing.\n\nFrance stress test and the rest are linked on the tracker: tigzig.com/private-credit.\n\n\"Market for lemons\"\n\nThat is the Bank of England's Sarah Breeden, deputy governor for financial stability, warning in April that private credit could tip into one.\n\nThe idea is Akerlof's, from a Nobel prize.. in a used car market the seller knows which cars are lemons and the buyer does not. So buyers discount everything, the good sellers walk away, and the bad cars are what is left. Information asymmetry, then adverse selection, then the market can collapse.\n\nPrivate credit has the same features. No market price.. the manager's own mark IS the price. Leverage stacked at borrower, fund and sponsor level, then sliced into structures so complex that only the people who built them can see through them. The 2008 CDO story ran on exactly this.\n\nAnd it is the same worry in different words everywhere: FSB on data gaps and valuation opacity, OCC on PIK masking credit deterioration, PIMCO now on a \"confidence gap\" between marks.\n\nBreeden's speech: bankofengland.co.uk.\n\nMy full read: tigzig.com/post/private-credit-market-for-lemons-may2026.\n\nAmar Harolikar · Decision Sciences & Applied AI · tigzig.com.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, TREMOR Macro Indicators, Quants Agent\n\nExplore: Private Credit & NDFI Stress Tracker, Analysis & Guides, MCP & Agent APIs\n\nMore posts: Private Credit and BDC Stress. Discounts at 13-Year Lows. Withdrawals Past the Gate. Regulators Now Arriving., Dimon Was Right. More Cockroaches Arrive. UK Lender MFS Collapses - $2.5B Across 12+ Global Banks., TREMOR. Six Free Tools for Banking and Insurance Analysts. Multi-Layer Validated. Built by a Practitioner."
},
{
"slug": "tigzig-quants-gpt-30-second-financial-analysis-custom-gpt",
"title": "TIGZIG Quants GPT: 30-Second Financial Analysis Custom GPT",
"tags": [
"custom-gpt",
"portfolio-analytics"
],
"content": "TIGZIG Quants GPT: 30-Second Financial Analysis Custom GPT\n\nImagine asking ChatGPT: 'Compare TCS, Gold, and Oil for me' ... and getting a full performance report back in seconds.\n\nQuick recap for those who missed it: This tool compares cross market securities - imagine Apple vs TCS vs Gold vs Nifty in seconds. Daily returns, drawdowns, CAGR, Sharpe ratios, CSV downloads - the works. All in one place. In 30 seconds or less.\n\nI just added a third way to use it: ChatGPT\n\nSame powerful engine. Same results. Same speed. But now with the familiar ChatGPT interface.\n\nWhy 3 versions?\nSuite - Fast, no-frills. Know Yahoo symbols - go here. Plus - full financials as CSV\nAgent - AI Smart Assist. Helps with Symbols. No rate limits (for now)\nChatGPT - For ChatGPT experience + available to free plan users also (with OpenAI rate limits)\n\nBoth overlap today...but agent gets exclusive features soon, and would be consolidated with Suite.\n\nThis is a build-as-you-go tool. Right now, CAGR is point-to-point (not a fan - too sensitive to endpoints). Rolling metrics & ML forecasts coming soon.\n\nReady to try?\n\nDirect Custom GPT link:\n\nUse the starter prompts - the GPT will guide you.\n\nSame 30-second turnaround. Same cross-asset comparisons. Now in ChatGPT.\n\nMore options:\n\nQuick reminder: I'm a CA, data scientist and an ex-derivative trader, not a portfolio manager. This supports my own research, not investment advice. Do your own due diligence always. All source code and methodology are open - hit the Docs button. Backend - FastAPI-MCP servers.\n\n\nRelated\n\nTools: Quants Suite, Quants Agent, Quants GPT\n\nExplore: Custom GPTs hub, Quant Apps hub, API and MCP directory\n\nMore posts: Three Tools I Built for Market Analysis - QSUITE, TREMOR, QREP, Automated Quant Reports with GPT: Run a stock, index, ETF, commodity, or crypto → get 3 formatted reports in minutes., Open Source Asset Comparison Tool: Compare Stocks, Indices, Crypto & Commodities in One Dashboard"
},
{
"slug": "tigzig-technical-analysis-ask-your-ai-aug2026",
"title": "Technical Analysis on Any Ticker Without Visiting My Site. Just Ask Your AI Agent.",
"tags": [
"technical-analysis",
"mcp",
"ai-coders"
],
"content": "Technical Analysis on Any Ticker Without Visiting My Site. Just Ask Your AI Agent.\n\nGet Technical analysis on any stock, index, crypto, commodity from TigZig Market Tools - without visiting my site. Paste this into Claude, ChatGPT or whichever agent you use:\n\n``\nget me a technical analysis report on nifty 50 from tigzig API\n``\n\nYou get back a nicely formatted PDF + Web report with technical charts, analysis and outlook. Any Yahoo ticker works, indices, stocks, FX. And you have a choice of models .. just ask your AI.\n\nTigZig is an AI Agent first site .. your AI figures out the rest on its own, where the API catalog is, which API to use, what the rate limits are.\n\nIt's open source, FastAPI with an MCP endpoint on the same app .. all docs, repos and links on site .. just ask your AI .. if it's a coder it can implement and customize this for you too.\n\nHow it works\n\nPrices come from Yahoo Finance on daily and weekly timeframes, indicators are computed with finta, EMAs, MACD, RSI and Bollinger Bands, and matplotlib draws the two charts. The numbers and the chart images both go to a vision model, which reads the chart and writes the interpretation. The markdown that comes back is converted to PDF and HTML by a second service built on ReportLab.\n\nSame app serves the web front end, the REST endpoint and the MCP endpoint, so a browser and an agent call the same code underneath.\n\nDocs page for technical analysis API and source: tigzig.com/apis/technical-analysis\n\nWhat this is, and what it is not\n\nThis is a reporting tool. The indicators are computed from Yahoo Finance data and the interpretation is written by a model reading the chart. It is not investment advice and it is not research.\n\nModels can misread a chart and data, and share a wrong interpretation .. very confidently. Always check the output.\n\nThe way I use this and other AI tools is to get the heavy lift done, get a different point of view, pull out missing angles .. but not standing in for my judgment.\n\nAnd I always validate outputs .. the numbers, formulas .. the deterministic part .. running it via more than one source. I am an analytics professional for over two decades, an ex-derivative trader .. I use AI heavily, from analysis to full stack tools. But I don't trust any output, from AI or human, unless it comes with tight validations .. interpretations are separate ..\n\ntigzig.com\n\n\nRelated\n\nTools: Quants Agent, QRep - Security Reports, TREMOR - Macro Stress Signals\n\nExplore: Technical analysis API docs, API + MCP hub, Claude for Analytics hub\n\nMore posts: Comparing Securities on Returns, Risk and Drawdowns. Four Routes into QuantStats, Depending on What Your AI Can Do., Tigzig is Built AI-Agent First. 300+ Assets, and the Fastest Way In Is to Ask Your AI., Every Tigzig MCP is now Streamable HTTP. Requesting existing users from LM Studio, ChatGPT, Claude etc to make a one time change if you have not already done so, Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users"
},
{
"slug": "tigzig-tremor-v2-mcp-330-indicators-jun2026",
"title": "Tigzig Tremor v2 Is Live. 330+ Macro + Credit Indicators Behind an MCP. 83% Lighter Startup. Just Ask Your AI Agent.",
"tags": [
"mcp",
"portfolio-quants"
],
"content": "Tigzig Tremor v2 Is Live. 330+ Macro + Credit Indicators Behind an MCP. 83% Lighter Startup. Just Ask Your AI Agent.\n\nPublished: 2026-06-24\n\nRelease announcement for Tigzig Tremor v2 - the MCP behind tigzig.com's 330+ macro and credit indicators across US, India, global macro, credit, markets and FX (17+ official publishers). Pulling macro data for analysis is mostly grunt work - finding the series, downloading, cleaning - so it now sits behind an MCP that AI agents (Claude Cowork, Claude in Excel, any agent) can ask in plain English. Sample asks in the post: 'US consumer net charge-offs, banks vs credit unions, last 15 years, chart it', 'S&P 500 and CAPE past 25 years, dual axis', 'Overlay Brent and WTI crude past 5 years'. The headline v2 change: catalog discovery split from tool calls. v1 stuffed the whole catalog into one ~8,000-token tool description the agent loaded up front; v2 surfaces an 8-category menu, the agent drills into only the tables it needs and pulls only the relevant bits - startup load cut from 7,854 to 1,352 tokens, 83% lighter. The discovery layer also carries house notes on each series (e.g. 'this charge-off is quarterly, multiply by 4 to annualize') so the AI does not quietly get it wrong. Free, open, no sign-up - use the live MCP, install your own, hit the APIs directly, or download the whole DB in your chosen format. Humans: tigzig.com -> TREMOR -> API / MCP. AI agents: api.tigzig.com.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/tigzig-tremor-v2-mcp-330-indicators-jun2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/tigzig-tremor-v2-mcp-330-indicators-jun2026 with header Accept: text/markdown\n\nTags\nmcp, portfolio-quants\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: APIs and MCP catalog, Quant Apps hub, Analysis archive\n\nMore posts: TREMOR - Macro Early Warning Dashboard - 50+ Indicators, Updated Daily, The Data Behind My Analyses Is Now Open. Download the Whole Database or Query a Single Row., Three Tools I Built for Market Analysis - QSUITE, TREMOR, QREP"
},
{
"slug": "tigzig-whole-database-downloads-aug2026",
"title": "Pulling Tigzig Data Through the API? There Is a Second Way In. The Whole Database Is Published for Download.",
"tags": [
"database-ai",
"infrastructure"
],
"content": "Pulling Tigzig Data Through the API? There Is a Second Way In. The Whole Database Is Published for Download.\n\nIf you are an analyst or a tool builder pulling Tigzig data through the API, there is a second way in. The whole database is published for download. Build your own micro-service, analyze it offline, or have your agent dig into it.\n\nThe dataset is refreshed back to back with the live one, and comes with a manifest with the generation timestamp on it, so you can see how current your copy is. Take a single table as parquet or CSV, or take the lot as one DuckDB or SQLite file.\n\nAPI or download? Sometimes the API is the right tool, and your API calls are welcome. In other cases, pulling the whole thing down and putting your own microservice in front of it gives you the same data without the API dependency and constraints.\n\nTip. If you have a Claude subscription it comes with Cowork, which comes with a python enabled sandbox installed on your local machine. It can download and analyze all the Tigzig data without breaking a sweat.\n\nFree, no signup, no API key.\nFor you: tigzig.com/apis\nFor your agent: api.tigzig.com\n\nWhat is in each\n\nMutual funds. NAV master from AMFI data, every scheme, going back to 2006. 37 million records.\n\nCricket. Ball by ball for men's Tests, ODIs and T20 internationals, 6,900+ matches from 2001, 3.8 million deliveries.\n\nGlobal macros. 300+ indicator series across the US, India, Europe and the OECD.\n\nIndia red flag. Related party transactions, insider trading, pledges, credit ratings and surveillance.\n\nEvery one is available table by table if you only want a slice, or as a single file with everything in it.\n\ntigzig.com/apis\n\nIf you are going to schedule this rather than pull it by hand\n\nRead the manifest first. Every dataset publishes one at /downloads/manifest, and it gives you the generation timestamp, the size of each file and the row counts, so a scheduled job can check the stamp and skip the download when nothing has changed.\n\nThat timestamp answers the question \"is this stale\". If it looks old, my live data is equally old and I am already looking at it, because the sync pipeline is monitored end to end.\n\nAnd if you would rather not read any of it, point your agent at api.tigzig.com and ask for the dataset you want.\n\nThe whole Tigzig database, published for download\n\n\nRelated\n\nTools: MFPRO - Mutual Fund Analytics, TREMOR - Macro Stress Signals, VIGIL - India Red Flag Events\n\nExplore: API and MCP catalog, Database AI hub, Claude for Analytics hub\n\nMore posts: The Cricket Download Files Now Sit on a Plain Path. Append the File Name and That Is the Whole URL., The Full Cricket Database Is Now Downloadable. Test, ODI and T20 Ball by Ball, in Parquet, CSV, DuckDB or SQLite., Pulling Indian Mutual Fund NAVs One Scheme at a Time? Three Ways to Get the Same Data in Far Fewer Calls., Getting a 429 From the MF NAV API? It Is Almost Never the Total. Two Changes Fix Nearly Every Case., Getting Errors on the Tigzig APIs? Is It You, or Is It Me? Now You Can Just Go and Look."
},
{
"slug": "track-trump-stock-trades-tool-release-may2026",
"title": "Track Trump's Stock Trades. Public Filing Explorer. 3,568 Trades, 986 Tickers, $220M-$750M. Searchable on TREMOR.",
"tags": [
"portfolio-quants"
],
"content": "Track Trump's Stock Trades. Public Filing Explorer. 3,568 Trades, 986 Tickers, $220M-$750M. Searchable on TREMOR.\n\nPublished: 2026-05-25\n\nTool release - Public Filing Explorer for Trump's stock trades, live on TIGZIG TREMOR. 3,568 equity trades from the latest OGE filing (Jan-March 2026), $220M-$750M disclosed value, all searchable across 986 tickers. Per-ticker price chart with buy/sell timeline overlay and news around each trade. Trades plotted on S&P 500 day by day. Raw OGE filings standardized with name-to-ticker mapping (filings come with messy long names, no tickers). CSV download for offline cuts. For 9 months before this filing it was mostly bonds; the May filing alone was almost all equities. Educational analysis, not financial advice.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/track-trump-stock-trades-tool-release-may2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/track-trump-stock-trades-tool-release-may2026 with header Accept: text/markdown\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nTigzig - Trump Trades Explorer\n\nTIGZIG · MARKET TOOLS\n\nAmar Harolikar\n\nPublic Filing Explorer\n\nTrack Trump's\nStock Trades.\n\nSearch every trade. Map it to price and news. Run your own cuts.\n\n3,568 equity trades, $220M-$750M in disclosed value.\n\nSearch every trade\n\n986 tickers, sort & filter\n\nName & ticker mapping\n\nraw filings standardized\n\nTrades on S&P 500\n\nday by day, buy vs sell\n\nPer-ticker price + news\n\ntrades on the company chart\n\nFull data download\n\nCSV, offline analysis\n\ntigzig.com\n→\nTREMOR\n→\nTRUMP TRADES\n\nAmar Harolikar\n\nDecision Sciences & Applied AI\n\ntigzig.com\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QRep - Security Reports, Quants Agent\n\nExplore: Quant Apps hub, Analysis, Data APIs and MCP\n\nMore posts: Trump Trades. From Mostly Bonds to 3,642 Equity Trades in One Filing. Free Tool to Search Every Transaction., Three Tools I Built for Market Analysis - QSUITE, TREMOR, QREP, TREMOR. Six Free Tools for Banking and Insurance Analysts. Multi-Layer Validated. Built by a Practitioner., S&P 500 Drawdown Analysis with QREP"
},
{
"slug": "tremor-ask-claude-agent-aug2026",
"title": "Reading Macro Charts on TREMOR? Send Them to the AI Agent Sitting There and Get a Read With the Latest News.",
"tags": [
"portfolio-quants",
"ai-coders"
],
"content": "Reading Macro Charts on TREMOR? Send Them to the AI Agent Sitting There and Get a Read With the Latest News.\n\nYou are analyzing macro charts on TREMOR - Indices, Delinquencies, Inflation, Commodities, Treasuries etc from the ~300 signals there. Want a read with the latest news. Just send it to the AI Agent sitting there.\n\n3 Options\nWhat changed this week sends every chart on your dashboard in one briefing.\nAnalyze selected charts takes up to five you pick and looks for trends, inflection points, correlations between them.\nDeep research does the same but goes out to newswires, ratings agencies and central bank releases, and cites what it finds.\n\nYou can edit the prompt before it goes, copy the analysis as markdown or print it as a PDF, and past runs stay in your browser local storage.\n\nFree, no sign up.\n\nTREMOR: tigzig.com/tremor ➜ Ask Claude Agent.\n\nIn the background\n\nEach chart goes to Claude as its data series, values and dates, along with the source label such as FRED or RBI DBIE, so the agent knows where the number came from and how much history it is looking at. 'What Changed This Week' sends the last 20 points per series, Analyze and Deep Research send the last 50.\n\nYou can also edit the prompt yourself before you send it. The three defaults are a starting point.\n\nA few caveats\n\nThis is AI commentary on the data, not a certified research note. It reads the numbers correctly and it searches for real news around them, but it can still miss context or misjudge how significant a move is. Treat it as a fast first pass, not the final word, and go to the primary source it cites before you rely on a statement.\n\nThe charts themselves are the same validated TREMOR series I publish everywhere else, FRED, RBI DBIE, MoSPI and the rest, so the numbers are good. What the agent does with them from there is its own read, and also very heavily influenced by the prompt and direction you provide. That is why I have kept the prompt customizable.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QRep - Security Reports, DATS-4 Database AI Suite\n\nExplore: Tremor API docs, Claude for Analytics hub, Analysis hub\n\nMore posts: TREMOR Opened With Sixteen Charts in a Fixed Grid. That Was My View. Now You Can Change All of It., Shiller CAPE Is at 41.2. In 145 Years the Only Period Higher Was the Dot-Com Peak., NY Fed Q2 2026 Household Debt. The Headline Delinquency Flow Fell, but Auto and Mortgage Went the Other Way., Point Your AI Agent at api.tigzig.com and It Works Out Which Dataset It Needs. 38 Million Rows Across India and the US, Free."
},
{
"slug": "tremor-bond-data-macro-fallout-sep2026",
"title": "All the Macro Data You Need to Track the Bond Market Rout, in One Place on TREMOR",
"tags": [
"portfolio-quants",
"database-ai"
],
"content": "All the Macro Data You Need to Track the Bond Market Rout, in One Place on TREMOR\n\nUS-10-YR touching 5.21% as I write. All the macro data you need to track the bond market rout, in one place on TREMOR. 340+ series from 25 sources, updated three times a day, covering bond yields, credit stress, jobs, inflation, oil, GDP and housing.\n\nBond yields are rising across the US, Europe and Japan. The US 10-year closed at 5.18% yesterday and touched 5.21% as I write. That kind of stress usually spreads into household credit, bank books, jobs and prices, and tracking it means pulling data from many agencies, each with its own site and release schedule.\n\nThere are many ways to use it\nInteractive tool - overlay series, switch to YoY or MoM, add a moving average of any period, add, drag and resize charts, and save any chart as a PNG.\nThe API and MCP server for your AI agent or scripting.\nFull database downloads as CSV, Parquet, DuckDB or SQLite.\n\nWhere to go\n\nFor you: TREMOR, tigzig.com/tremor\n\nFor your AI agent: api.tigzig.com\n\nThe bond market rout, from earlier today: The bond market rout\n\nAll my earlier analyses: tigzig.com/analysis\n\nSome of my earlier analyses on this\nThe bond market rout, 25 September. Most of the rise in the US 10-year is the real yield, with expected inflation at about where it was one and three years ago. The bond market rout\nBond markets are sounding an alarm, 6 September. The US 30-year was at 5.25 per cent while the Fed was still cutting, and Europe and Japan were moving the same way. Bond markets are sounding an alarm\nThe US job market: the calm before the storm? 24 September. The headline numbers look fine, but job growth has slowed to 0.4 per cent a year and hiring is near its lowest on record. The US job market: the calm before the storm?\nAI valuations, a correction looks more likely, 31 August. A correction looks more likely, and it rests on free cash flow arriving from 2028. A correction in AI valuations\n\nAnalysis hub\n\nTREMOR: 340+ macro and rate series from 25 sources, updated three times a day, covering bond yields, credit stress, jobs, inflation, oil, GDP and housing"
},
{
"slug": "tremor-chart-layout-aug2026",
"title": "TREMOR Opened With Sixteen Charts in a Fixed Grid. That Was My View. Now You Can Change All of It.",
"tags": [
"portfolio-quants",
"database-ai"
],
"content": "TREMOR Opened With Sixteen Charts in a Fixed Grid. That Was My View. Now You Can Change All of It.\n\nTREMOR is my macro dashboard on tigzig.com. It opens with sixteen charts in a fixed grid. That was my view, and if it is not yours, you can now change all of it.\nDon't want the BSE Sensex chart - delete it\nWant Unemployment and Delinquency side by side - drag it\nWant to replace Brent with Gold - overwrite it\nCharts look tiny - expand it\nToo cluttered - resize it\nWant to look at 10yr & 30yr treasury on same chart - overlay it\nAdvance retail sales - need % change - YOY / MOM it\nChart too noisy - moving average it - with whatever period you want\nNeed a new chart - add it - from ~300 indicators\nWant to paste to deck - png it\nWant to check data - copy it\nWant an AI analysis of charts + latest news - AI it\nWant to expand to 30 years - change it\nDon't want to see labels - hide it\nWant the full database - download it\n\nDon't want to analyze it yourself? Then just point your AI to the API docs and have it download and analyze whichever series you want.\n\nFor you\nTool: tigzig.com/tremor\nAPI: tigzig.com/apis\n\nFor your AI\n\n``\ngo to api.tigzig.com, hit the tremor API and let me know what it has\n``\n\nFree. No sign up.\n\nWhat is actually in there\n\nAround 330 series, ~250 US and ~50 India.\n\nA big chunk of it is credit. FDIC bank aggregates, so loan balances, delinquency, charge-offs. NCUA for credit unions. NY Fed household debt, 30+ and 90+ delinquency split by card, auto and mortgage.\n\nThen the usual macro off FRED, rates, labour, inflation, GDP, housing. India is MoSPI and RBI, CPI, IIP, WPI, GDP, bank credit and deposits, and Nifty PE from NSE.\n\nPlus odds and ends like OECD business confidence, Manheim used vehicle prices, Shiller, European bond yields from ECB and Bundesbank.\n\nHeavily documented as very often I will keep forgetting how the series has been pulled, so every series has a section in the docs saying what it is and where it is pulled from. And on the chart itself there is an info button, click it and you get what the series measures, the source, and how it is normally read. So you can see what the series is about right when you are looking at the chart, that's the way I like it.\n\nOn the AI side you have two ways to go\n\nSimplest, just give your AI the API. No install, no MCP, nothing to set up. It reads the docs, works out the endpoints and pulls what it needs.\n\nIf you would rather work through MCP, there is an MCP server for it. Or take the API and wrap your own thin MCP around it. Either way works.\n\nFull data download is there too. Go to the data and API section in the app and pull the whole thing down, or have your agent do it.\n\nLooking at my logs there is a fair bit of downloading going on, so my guess is plenty of people would just rather have the file and do the analysis in their own tools. That is what it is there for.\n\ntigzig.com/apis\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QRep - Security Reports, DATS-4 Database AI Suite\n\nExplore: Tremor API docs, API and MCP catalog, Analysis hub\n\nMore posts: Shiller CAPE Is at 41.2. In 145 Years the Only Period Higher Was the Dot-Com Peak., NY Fed Q2 2026 Household Debt. The Headline Delinquency Flow Fell, but Auto and Mortgage Went the Other Way., Point Your AI Agent at api.tigzig.com and It Works Out Which Dataset It Needs. 38 Million Rows Across India and the US, Free., Q1 2026 US Lending Data Live on Tigzig. NCUA + FDIC + NY Fed in One API. Two of Three Not on FRED. AI-Agent Ready."
},
{
"slug": "tremor-claude-managed-agents-macro-analysis",
"title": "Anthropic Released Claude Managed Agents Yesterday, Live on TREMOR Today",
"tags": [
"portfolio-quants",
"ai-coders"
],
"content": "Anthropic Released Claude Managed Agents Yesterday, Live on TREMOR Today\n\nPublished: 2026-04-09\n\nClaude Managed Agents integrated into TREMOR macro dashboard. AI analysis on 70+ indicators across US and India. Three modes: full dashboard briefing, selected chart analysis, deep research with citations. Covers equities, bonds, gold, oil, forex, unemployment, delinquencies, valuations, inflation.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/tremor-claude-managed-agents-macro-analysis\nMarkdown of the HTML page: send GET https://tigzig.com/post/tremor-claude-managed-agents-macro-analysis with header Accept: text/markdown\n\nTags\nportfolio-quants, ai-coders\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: Quant Apps hub, Analysis hub, APIs and MCP\n\nMore posts: Tigzig Tremor v2 Is Live. 330+ Macro + Credit Indicators Behind an MCP. 83% Lighter Startup. Just Ask Your AI Agent., TREMOR - Macro Early Warning Dashboard - 50+ Indicators, Updated Daily, Three Tools I Built for Market Analysis - QSUITE, TREMOR, QREP, Claude Cowork and Claude Code Have Quietly Become My Universal Front End. More Than Excel, Jupyter or Colab."
},
{
"slug": "tremor-macro-api-ask-your-ai-jul2026",
"title": "Tremor: A Free, No-Auth API for 330+ Macro and Market Indicators, 350K records, Refreshed 3x/Day. Ask your AI to explore.",
"tags": [
"mcp",
"infrastructure"
],
"content": "Tremor: A Free, No-Auth API for 330+ Macro and Market Indicators, 350K records, Refreshed 3x/Day. Ask your AI to explore.\n\nPublished: 2026-07-12\n\nRelease note on the current state of Tremor, tigzig.com's free open API and MCP server for macro, credit, valuation and market data. 331 public indicators across 8 categories, 350K records, refreshed 3x a day, no key and no sign-up. Ask-your-AI framing - point an AI agent at api.tigzig.com's Tremor and it reads the category menu (8 high-level buckets), drills into the one that matches the question, gets the full indicator list with metadata, searches if needed, then pulls the series - one prompt, the human never touches the docs. MCP also available for agents that prefer that surface. Category breakdown: US macro (35 - FRED 32 + Michigan Consumer Sentiment 1 + Manheim Used Vehicle Value Index 2), US credit stress (216 - NY Fed Household Debt 64 + FDIC bank aggregates 80 + NCUA credit-union aggregates 72, covering delinquency by product / charge-offs / originations by credit tier), US valuation (1 series - Shiller CAPE since 1881, with the usvaluation grouping also surfacing the Buffett indicator and nominal GDP from usmacro and the 10Y Treasury yield), US insurance (currently empty - Fed Z.1 Life + P&C invested-asset breakdown scaffolded and loading), India macro and credit (47 - MoSPI 27 + RBI 15 + NSE 3 + FRED USDINR 1 + one computed CD ratio), Global macro (8 - OECD Business Confidence for 4 countries + Bundesbank 10Y Bund + Bank of England 10Y Gilt + ECB Euro-area 10Y yield curves for AAA and all-issuers), Currencies (7 daily FX pairs including USDINR + EM Asia + Fragile-X peer cuts USDJPY / USDKRW / USDTWD / USDPHP / USDTRY / USDVND), Markets (16 daily Yahoo Finance symbols - S&P 500, Nifty 50, Nikkei 225, Hang Seng, DAX, Shanghai, US 10Y ^TNX, DXY, gold, silver, copper, WTI, Brent, natural gas, USDINR, USDJPY). Each indicator ships a notes field with the gotchas baked in for the agent: usrealgdp is already SAAR-annualized (do not multiply by 4), innifty50pe has a standalone-to-consolidated methodology break in April 2021, innetfdi negative means outflow and there is a 6+ month publication lag. Pipeline detail: pull via API where available (FRED, RBI WSS, ECB Statistical Data Warehouse, OECD, Bundesbank, Bank of England); for sources published only as files with year-to-year format drift (NCUA 5300 Call Reports, FDIC SDI aggregates, NY Fed Household Debt Study, MoSPI eSankhyiki, RBI DBIE Excel, Shiller's ie_data.xls, Manheim MUVVI) the file is downloaded, parsed with error correction, standardized into a series. Runs 3x a day - Fed publishes at 3pm ET, we pick it up in the next cycle and it is live in Tremor by early evening. Multi-cadence handling: RBI is a mix (weekly rates + fortnightly credit + monthly trade/FDI/FPI), MoSPI monthly or quarterly by series, NSE Nifty PE/PB daily, quarterly GDP + GFCF + credit-union aggregates land once a quarter and get ingested the day published. Pipeline does error checking: sums validate, YoY consistency checked, data quality issues (missing values, spikes that do not match the source document) flagged in metadata so the caller sees them. Sources summary lists FRED, NY Fed, FDIC, NCUA, RBI, MoSPI, NSE, ECB, Bundesbank, BoE, OECD, Shiller, Michigan, Manheim, Yahoo Finance with a one-line note on each. Includes 'what moved recently' section with current-through dates on Treasury curve + VIX + Baa credit spread + 10Y TIPS + 10Y-2Y spread + 30Y fixed mortgage rate + PMMS + initial jobless claims + Bund + Gilt + Euro-area yield curves + the 16 daily market symbols. Launched March with ~50 indicators; now 331 and growing. Single-image release post with the one-pager PNG at the bottom.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/tremor-macro-api-ask-your-ai-jul2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/tremor-macro-api-ask-your-ai-jul2026 with header Accept: text/markdown\n\nTags\nmcp, infrastructure\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Full API + MCP catalog, Market tools hub, Analysis + guides\n\nMore posts: Tigzig Tremor v2 Is Live. 330+ Macro + Credit Indicators Behind an MCP. 83% Lighter Startup. Just Ask Your AI Agent., FRED, 845,000 US Economic Series - Your AI Can Pull Any of Them Directly. No MCP, No Plugin, No Install., Tigzig APIs - Reading the Logs, Shipping the Fixes. Status Page Live, R2 Downloads, Self-Correcting 4xxs., TREMOR. Six Free Tools for Banking and Insurance Analysts. Multi-Layer Validated. Built by a Practitioner."
},
{
"slug": "tremor-macro-dashboard-50-indicators",
"title": "TREMOR - Macro Early Warning Dashboard - 50+ Indicators, Updated Daily",
"tags": [
"portfolio-quants"
],
"content": "TREMOR - Macro Early Warning Dashboard - 50+ Indicators, Updated Daily\n\nPublished: 2026-03-27\n\nTREMOR macro early warning dashboard with 50+ indicators for US and India. Updated daily from FRED, RBI, NSE, MOSPI, NY Fed. Swap charts as news breaks, YoY/MoM/MA toggles, PNG export to PowerPoint, CSV export, news feed from 30+ sources. Free, no login.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/tremor-macro-dashboard-50-indicators\nMarkdown of the HTML page: send GET https://tigzig.com/post/tremor-macro-dashboard-50-indicators with header Accept: text/markdown\n\nTags\nportfolio-quants\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: Quant apps hub, Analysis deck, Guides hub\n\nMore posts: Tremor - Macro Early Warning Signals - March 2026, Tigzig Tremor v2 Is Live. 330+ Macro + Credit Indicators Behind an MCP. 83% Lighter Startup. Just Ask Your AI Agent., TREMOR. Six Free Tools for Banking and Insurance Analysts. Multi-Layer Validated. Built by a Practitioner., Three Tools I Built for Market Analysis - QSUITE, TREMOR, QREP, Are We Headed for Stagflation-Lite"
},
{
"slug": "tremor-macro-signals-march-2026",
"title": "Tremor - Macro Early Warning Signals - March 2026",
"tags": [
"portfolio-quants"
],
"content": "Tremor - Macro Early Warning Signals - March 2026\n\nPublished: 2026-03-14\n\nS&P down 5%, Nifty in bear territory down 12% from highs. CAPE at 40-year highs, gold going vertical, crude shooting towards 2008 levels, US delinquencies crossing COVID highs. When macro signals move together - tremors before a quake. 12 chart slides from tremor.tigzig.com.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/tremor-macro-signals-march-2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/tremor-macro-signals-march-2026 with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/TREMORMAR2026.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nWhere are we headed\n\nBond yields spiking\n\nConsumer sentiment at one of lowest in past 20 years\n\nOil Remains high\n\nUS Q425 GDP growth revised downwards to 0.7%\n\nUS Unemployment & Non-Farm Payrolls flashing red\n\nBond markets pricing in high inflation expectation\n\nUS Past 20 Years\n\nPast 6 Months\n\nUS Treasuries 10Y Yield\n\nGermany 10Y Bund Yield\n\nInstitutional and bond desk expecting higher inflation. Which make the current yields unattractive, causing sell-off and consequential rise in yields. Becomes self fulfilling – the expectation itself create outcome in bond markets increasing risk to real economy CNBC – Rising European Yields, Bloomberg Podcast\n\nUS Feb’26 PPI annual inflation at 3.4%, higher than expected\n\nUS Past 20 Years\n\nThe producer price index is a measure of pipeline costs that producers receive for their products and an early indicator of consumer inflation. The annual inflation rate was at 3.4%, highest in past year. And the oil shock is yet to be baked in\n\nConsumer sentiments running abysmally low\n\nMichigan Consumer Sentiment results released March 27 came in at 53.3 amongst the lowest reading (7th lowest) in past 20 years\n\nUS GDP growth rate revised to 0.7%\n\nThe first revision of GDP on13th March 2026 was a 50% lower than the previous 1.4% and below Dow Jones consensus forecast of 1.5% CNBC (Seasonally & inflation adjusted Annualized QoQ growth rate)\n\nOil, unemployment and delinquencies in red zone\n\nBrent continues to be above $100 and even when the Iran war ends and prices fall, the transmission won’t be instant. Some capacities may get online in months, others might take years. All while pressures continue on employment side and rising delinquencies.\n\nEconomists raising odds of a recession\n\nMoody’s Analytics’ model has raised its recession outlook for the next 12 months to 48.6%. Goldman Sachs boosted its estimate to 30%. Wilmington Trust has the odds at 45%, while EY Parthenon has it at 40%, with the caveat that “those odds could rapidly rise in the event of a more prolonged or severe Middle East conflict.”\n\nIn normal times, the risk for a recession in any given 12-month span is around 20%. So while the current predictions are hardly certainties, they signify elevated risk.\n\n“I’m concerned recession risks are uncomfortably high and on the rise,” said Mark Zandi, chief economist at Moody’s Analytics. “Recession is a real threat here.”\n\nReports CNBC March 25, 2026\n\nUS Shiller CAPE (Warren Buffett Indicator) is at its 2nd highest level ever and inching towards the highest seen during the dot com bubble\n\nGold touching new highs and orbiting the outer world\n\nPast 20 Years\n\nCrude nearly doubled since December and rapidly going towards 20 year high touched last time in 2008\n\nUS advance retail sales growth is on downward trajectory\n\nUS 30+ DPD has already crossed the COVID highs\n\nPast 20 Years\n\nUS Unemployment creeping up steadily the past 2 years\n\nPast 20 Years\n\nUS 10Y-2Y treasury spreads are coming off its highs\n\nAs many forecasts as there are expert. But mine is always 100% correct.Market goes up, down, or sideways.I don't trade actively anymore.But the trader's mindset stays.Have a setup ready for each An Entry. An Exit. A Stop loss.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: Download the Tremor deck (PDF), Market tools hub, Analysis hub\n\nMore posts: Are We Headed for Stagflation-Lite, Three Red Flags Hit Together. 30Y at 5.14%, PPI at 5.99%, Brent Above $100. Setup Worse Than 2000 or 2007., TREMOR - Macro Early Warning Dashboard - 50+ Indicators, Updated Daily, S&P at All-Time High. Under the Hood, Things Are Cracking. Just Like October 2007."
},
{
"slug": "tremor-monthly-update-aug2026",
"title": "Shiller CAPE Is at 41.2. In 145 Years the Only Period Higher Was the Dot-Com Peak.",
"tags": [
"portfolio-quants",
"database-ai"
],
"content": "Shiller CAPE Is at 41.2. In 145 Years the Only Period Higher Was the Dot-Com Peak.\n\nShiller CAPE is at 41.2. In 145 years of history the only period higher was the dot-com peak. That is one of the readings from this month's Tremor update.\n\nThe US 30-year Treasury is at 5.23%, and 2026 is the first year since 2007 that the long bond has been above 5.2%. US auto loans are running a 90+ day delinquency flow of 3.00%, the highest since the first quarter of 2010. Brent is back at 95 dollars, after peaking above 138 in April and falling to around 70 by the end of June.\n\nOn the data side, India bank credit now goes back to February 2017 rather than December 2022, which gives 113 months of sector-wise history across 33 sectors. NY Fed household debt for Q2 2026 is loaded across 64 series, on the tool and on the public API. Daily prices cover the major equity indices along with Brent, natural gas, copper and gold, and BSE Sensex now goes back to 1997.\n\nFree, no signup and no key.\n\nInteractive tool, advanced charts, API & MCP, full database download.\n\nGo to tigzig.com, then TREMOR.\n\nIf you would rather have an agent pull it, the same data is on the API and over MCP: tigzig.com/apis/tremor\n\nIf you would rather point an agent at it\n\ntigzig.com/tremor\n\nIf you would rather point an agent at this than click through charts, api.tigzig.com takes a plain question and works out which dataset it needs, across India and US macro, credit, bank aggregates and market prices, and builds out the charts and analysis for you.\n\nRelated posts from this month:\nPoint Your AI Agent at api.tigzig.com and It Works Out Which Dataset It Needs\nNY Fed Q2 2026 Household Debt, where the headline delinquency flow fell but auto and mortgage went the other way\n\nGuides on working with AI to build your analysis, charts and reports\nClaude Cowork: Working with Claude Cowork and Claude Code\nChatGPT Work: ChatGPT Work comes with your Go or Pro subscription\nGoogle Antigravity: Antigravity, a full AI coding agent\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QRep - Security Reports, DATS-4 Database AI Suite\n\nExplore: Tremor API docs, Analysis hub, Claude for Analytics hub\n\nMore posts: NY Fed Q2 2026 Household Debt. The Headline Delinquency Flow Fell, but Auto and Mortgage Went the Other Way., Point Your AI Agent at api.tigzig.com and It Works Out Which Dataset It Needs. 38 Million Rows Across India and the US, Free., Q1 2026 US Lending Data Live on Tigzig. NCUA + FDIC + NY Fed in One API. Two of Three Not on FRED. AI-Agent Ready., Auto Losses Past the GFC Peak. Cards Almost There. Credit Unions Already at GFC Loss Rates. NY Fed + NCUA Data Now on Tigzig."
},
{
"slug": "tremor-news-jev-rank-oct2026",
"title": "Jev Rank on TREMOR News Scores Every Headline Against What You Want to Read",
"tags": [
"portfolio-quants",
"ai-coders"
],
"content": "Jev Rank on TREMOR News Scores Every Headline Against What You Want to Read\n\nI added Jev Rank to the news section on TREMOR. It scores every headline from 0 to 100 against what you want to read, and sorts the page by that score. About 800 headlines from 64 feeds, ranked in under 2 seconds.\n\nThere are three ways to use it.\nRun Jev Rank: scores the headlines against your interests, a default set until you change it.\nEdit interests: lets you set your own topics and example headlines, saved in your browser. The detailed one.\nFocus box: takes a topic, or a few sentences in plain words, typed or dictated, and ranks against that instead. For quick stuff.\n\nI use the focus box for quick stuff ... and the Edit interests for creating more permanent interest profiles .. multiple profiles, saved as files, that I load based on the work I am doing.\n\nJev is TypeSafe's new System One model. You give it some context and a set of questions, and each answer comes back as numbers. A question can ask it to pick one option from a list, to rate something against levels you describe, or to say how likely a statement is to be true. I use the last kind: one yes or no question per headline, would this reader want to read it, and the probability of yes becomes the score.\n\nIt is free to use. No signup.\n\nhttps://www.tigzig.com/tremor ➜ News.\n\nWhat it covers and how to work with it\nSources: 64 feeds from about 50 publishers, including Reuters, Bloomberg, Financial Times, CNBC, AP, BBC, Economic Times, LiveMint, Business Standard, the Fed, ECB, BIS and IMF.\nSections: US Markets, Global Finance, India, World, Central Banks, Research, Tech, AI and Tech News, and Cyber Security. Jev ranks whichever section and time window you are on.\nFocus box: it replaces your interests for that ranking. Back to my interests returns to them in one click. Jev reads the words literally, so a specific focus gives a tighter list.\nScores: kept for the session. Switching between your interests and a focus you already ran is instant.\nRefresh: pulls every feed again. Only the new headlines get scored, and the rest keep their scores. Reloading the browser tab clears the scores, so press Run Jev Rank again.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QDesk - Quant Report Desk, Quant Apps\n\nExplore: Analysis hub, TREMOR API docs, What is new\n\nMore posts: Reading Macro Charts on TREMOR? Send Them to the AI Agent Sitting There and Get a Read With the Latest News., TREMOR Opened With Sixteen Charts in a Fixed Grid. That Was My View. Now You Can Change All of It., Tremor: A Free, No-Auth API for 330+ Macro and Market Indicators, 350K records, Refreshed 3x/Day. Ask your AI to explore."
},
{
"slug": "tremor-six-tools-banking-insurance-analysts-may2026",
"title": "TREMOR. Six Free Tools for Banking and Insurance Analysts. Multi-Layer Validated. Built by a Practitioner.",
"tags": [
"portfolio-quants"
],
"content": "TREMOR. Six Free Tools for Banking and Insurance Analysts. Multi-Layer Validated. Built by a Practitioner.\n\nPublished: 2026-05-22\n\nSix free tools for banking and insurance analysts on TREMOR (tigzig.com). Bank Aggregates from FDIC, Credit Union Aggregates from NCUA, Consumer Credit from NY Fed, Bank-Level NDFI from FFIEC, Insurance Assets from Fed Z.1, Global Macro Indicators from multiple sources. Primary sources only - no third-party aggregators. Every tool carries a three-layer Validations tab. Walks through what it took to validate the data: regulator nuances (NCUA YTD charge-offs need de-cumulation, NY Fed revises back-data, FDIC late-filers drift against QBP totals), abnormal patterns, and even an apparent error in the regulator's own filing. Built by an analyst, for analysts. Free. No ads. No login.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/tremor-six-tools-banking-insurance-analysts-may2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/tremor-six-tools-banking-insurance-analysts-may2026 with header Accept: text/markdown\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nTIGZIG · US Analyst Tools · 19 May 2026\n\nLive data. Documented sourcing. Validated. Free.\n\nRun live charts, cuts and analytics → tigzig.com → TREMOR\n\nSix tools, one place\n\n| Tool | Source |\n|---|---|\n| Bank Aggregates | FDIC |\n| Credit Union Aggregates | NCUA |\n| Consumer Credit | NY Fed |\n| Bank-Level NDFI | FFIEC |\n| Insurance Assets | Fed Z.1 |\n| Global Macro Indicators | FRED + OECD |\n\nPlus, on every tool\nMethodology\nValidations\nData Download\nCustomizable Charts\nSeries + Moving-Average Overlays\nNews from 30+ Sources\n\nAmar Harolikar · Decision Sciences & Applied AI · tigzig.com\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, TREMOR Macro Indicators, Quants Agent\n\nExplore: Private Credit and NDFI Tracker, Quant Apps hub, Analysis library\n\nMore posts: Releasing US-CREDIT. For Analysts Tracking US Banking. Interactive Tool Covering FDIC SDI and NY Fed CCP., US Insurance Assets Tracker - 25 Years of Data, 73 Series, Life & P&C, NDFI Analytics - US Bank Exposure to Non-Bank Lenders, US Credit Unions on TREMOR. Card Charge-Offs at 5.11% - Already Past the 2009 GFC Peak."
},
{
"slug": "trump-2025-annual-disclosure-jul2026",
"title": "Trump's 2025 annual financial disclosure: all 28,007 records, extracted and searchable",
"tags": [
"portfolio-quants"
],
"content": "Trump's 2025 annual financial disclosure: all 28,007 records, extracted and searchable\n\nPublished: 2026-07-11\n\nRelease post announcing that OGE certified and released President Trump's 2025 annual public financial disclosure (Form 278e, 927 pages) on June 30, 2026, and the whole thing has been extracted into a structured, searchable, downloadable dataset - 28,007 records across eight tables - and added to the Trump Trades tool on tigzig alongside the periodic 278-T trade reports. What is now on the tool: an Overview landing page explaining the two filing types, an At-a-glance dashboard, a Companies tab with ticker rollup, a vs S&P 500 tab that plots all 142 trading days on the index chart, full table viewers with search / filters / sort / pagination for Transactions / Holdings / Business & Income, a Liabilities & Gifts tab, a Data tab with record counts + per-table CSVs + the full JSON + the source PDF, and an About & Validation tab. Extraction breakdown: Transactions Part 7 = 21,285; Holdings Part 6 = 6,181; Business entities Part 2 Schedule 1 = 501; Liabilities Part 8 = 11; Gifts Part 9 = 11; Spouse assets and income Part 5 = 11; Positions Part 1 = 4; Agreements Part 3 = 3. Excludes the 54-page Exhibit A trademark schedule (unstructured) and Part 4 (filed not applicable). Non-obvious rule flagged in the post: the annual's transaction section does NOT repeat trades already disclosed on the 278-T trade reports (OGE rule, and this filing uses it), so the 21,285 trades in the annual are ones the public had not seen before (mostly from the managed stock accounts), while the municipal and corporate bond trades are only on the 278-Ts - neither document alone is the full 2025 trading record, the union is; empirically the two sets are essentially disjoint, and the filing's own cover notes that late filing fees were paid for transactions not previously reported on 278-Ts. Trading detail: 15,524 purchases against 5,761 sales, midpoint ~$942M bought vs $286M sold (net buyers all year), across 8 managed accounts (uneven - Account 7 alone holds 10,311 trades, Account 8 has 3,972, Account 6 has 2,130, Account 1 has 1,757), 2,148 distinct securities, 51% of trades in the $1,001-$15,000 disclosure band and 83% under $50,000 (pattern of professionally-managed direct-indexing / diversified accounts, not concentrated stock-picking), trades on 142 trading days including net buying near the bottom of the April 2025 drawdown. Top securities by gross midpoint volume with tickers consolidated: Microsoft (~$43.7M bought / $1.7M sold, 84 trades), NVIDIA ($42.0M bought, 65 trades), Apple ($41.4M bought, 71 trades), Amazon ($21.6M bought, 74 trades), Alphabet ($18.5M bought, 93 trades); by trade count Alphabet 93 / Meta 86 / Microsoft 84 / Amazon 74 / Apple 71; money-market sweeps excluded. Business income: 97 exact-dollar business lines totaling $2.16B - Celebration Coins royalties $635,068,835, World Liberty Financial token sales $526.8M across token-sale lines (main $236.25M + per-wallet distributions of $150.6M ETH / $56.0M USDC / $42.3M USD / $33.5M BTC), stablecoin-related DT Marks SC LLC $196.9M, Trump Endeavor 12 (Doral golf resort) $121.9M, Mar-a-Lago Club $77.5M, WLF Holdco equity sale $65.6M. Litigation settlements $86.5M - Twitter/X $8M, ABC $16M, CBS $16M, Meta $24.5M, YouTube $22M. Melania Trump entities separately $17.2M exact license proceeds in Part 5 (film $10.71M + NFTs $6.01M + book $0.52M). Note that press coverage generally aggregates crypto items to ~$1.4B total but the tool lists each line as filed rather than as one combined number. Holdings: 6,181 lines totaling ~$1.86B midpoint across the 8 investment accounts + a revocable trust + a family trust + direct holdings - heavy in municipal bonds and blue-chip equities. Revocable trust holds a ladder of brokerage money-market accounts including two reported at 'Over $50,000,000' each, and the direct holdings include a line reading 'INVESTMENT IN GOLD BARS' in the $500K-$1M band. Liabilities: 11 lines including the two E. Jean Carroll judgments (one over $50M stayed pending appeal + bonded), the NY AG judgment noted as overturned on appeal, a Charles Schwab pledged-asset line over $50M opened in 2025, and several mortgages three of which are marked paid off in 2025. Gifts: 11 gifts totaling $372,050 - mostly event tickets (Super Bowl, FIFA World Cup Final, Ryder Cup, UFC) plus a $250,000 sculpture. Convention flagged: securities values and trade amounts are disclosed as ranges so every dollar total involving securities uses the midpoint of each range and 'Over $X' ranges use the lower bound - treat those totals as approximations; business income figures are exact dollar amounts as filed. Validation methodology on the tool: (1) two independent extraction methods with unrelated parsing engines reconciled field-by-field across every row to 100% agreement, (2) completeness gates on the filing's own row numbering + every amount / type / date matched against the official OGE vocabulary, (3) independent AI vision cross-reads on sampled pages + the full business-entity schedule blind to the dataset with disagreements adjudicated against a high-resolution page image (zero extraction errors found), (4) external tallies checked against reported figures (exact-dollar match on the $635M Celebration Coins line). Filing quirks preserved as filed (inconsistent capitalization, occasional typos, filer-truncated names). Author flags plainly that OGE publishes the filing only as a PDF with no machine-readable version, and this dataset is a best-effort reproduction of a 927-page government document not the document itself - the source PDF is one click away on the tool for verification. Sources: the OGE-certified 278e PDF (927 pages) and the OGE release note of June 30, 2026. Analytical release post - single-image, one-pager PNG at the bottom.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/trump-2025-annual-disclosure-jul2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/trump-2025-annual-disclosure-jul2026 with header Accept: text/markdown\n\nTags\nportfolio-quants\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Analysis + guides, Full API + MCP catalog, Market tools hub\n\nMore posts: Trump Trades. From Mostly Bonds to 3,642 Equity Trades in One Filing. Free Tool to Search Every Transaction., Track Trump's Stock Trades. Public Filing Explorer. 3,568 Trades, 986 Tickers, $220M-$750M. Searchable on TREMOR., Why Trump Is Going After Jamie Dimon Right Now. Two Fronts - Debanking + Crypto. The DOJ Subpoena Made It Real., TREMOR. Six Free Tools for Banking and Insurance Analysts. Multi-Layer Validated. Built by a Practitioner."
},
{
"slug": "trump-trades-3642-equity-trades-tool-may2026",
"title": "Trump Trades. From Mostly Bonds to 3,642 Equity Trades in One Filing. Free Tool to Search Every Transaction.",
"tags": [
"portfolio-quants"
],
"content": "Trump Trades. From Mostly Bonds to 3,642 Equity Trades in One Filing. Free Tool to Search Every Transaction.\n\nPublished: 2026-05-21\n\n9 months of mostly bonds in President Trump's OGE filings broken by one filing on 14 May 2026 carrying 3,642 equity trades, $220M to $750M. Heavy activity in Nvidia, Apple, Microsoft, Amazon, Meta, Oracle. 15 filings since Aug 2025 - earlier 14 combined had 1,726 trades almost all bonds. Built a free interactive tool on top of the OGE PDFs - every transaction searchable across 1,058 tickers, click any company to see trade timeline overlaid on price chart, CSV download. Short analysis deck included on most active tickers, top buys, top sells, and policy overlaps press has flagged.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/trump-trades-3642-equity-trades-tool-may2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/trump-trades-3642-equity-trades-tool-may2026 with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/TRUMPTRADESMAY2026_C.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nAnalysis of Q1 2026 Stock-Trading Disclosures - 20 May 2026\n\nTool · Analysis · Data · News\n\n3,642 trades. $220M - $750M disclosed range. For 9 months President Trump's filings showed mostly bonds. Then on 14 May 2026 a single OGE filing landed with 3,642 equity trades. This brief lays out what the filing shows: filing-vs-filing composition, the most active tickers, the largest buys and sells, and the trade-vs-action timing the press has flagged. A free interactive tool covering every trade is live on TREMOR.\n\nThe 14 May Filing at a Glance\n\nTransaction Report covering 06 Jan to 27 Mar 2026 - 45 trading days\n\n3,642 equity trades filed in one report. ~58 per market day.\n2,346 buys / 1,296 sells. 1.8 buys per sale.\n52 buys + 29 sells per day on average.\n1,058 distinct tickers. Roughly 3.4 trades each.\n$220M-$750M disclosed range (OGE Form 278-T reports each trade as a range bucket; total is the sum of those ranges).\n\nThe May Filing Dwarfs Every Prior One\n\n15 OGE filings since Aug 2025. The May 14 filing alone has more trades than the previous 14 combined.\n\nPrior 14 filings: 1,726 trades. Almost all muni and corporate bonds.\n14 May filing: 3,642 trades. Almost all equities.\n\nTrades per filing (most recent first):\nMay '26 - 3,642\nApr '26 - 69\nMar '26 - 175 / 174 / 83 (three filings)\nJan '26 - 191\nDec '25 - 108\nNov '25 - 32 / 39 / 79 / 57 (four filings)\nSep '25 - 25\nAug '25 - 149 / 507 / 38 (three filings)\n\nThe shift is visible at a glance: the 14 May report alone exceeds the cumulative trade count from August 2025 through April 2026.\n\nThe Interactive Tool\n\nBuilt on top of the raw OGE PDF filings. Live on tigzig.com → TREMOR → Trump Trades.\n\nFour views inside the tool:\nFilings tab. Composition before-vs-after for all 15 OGE filings.\nCompanies tab. 1,058 tickers searchable. Sort, filter, drill, export.\nLive price + news. Trade timeline overlaid on the price chart + recent news for any ticker.\nS&P 500 view. Trades plotted against the index day by day.\n\nCSV download for offline analysis.\n\nHow to Read the Amounts in This Deck and the Tool\n\nOGE discloses each trade in a range, not an exact value.\n\nOGE Form 278-T reports each trade as a range bucket (e.g. $250K-$500K), not an exact figure. Every dollar total in this deck and the tool is the sum of the midpoints of those buckets.\nWhat this CAN tell you: general scale and direction of activity.\nWhat this CANNOT tell you: the exact dollar value of any single trade, or a precise total in actual dollars.\n\nSource: OGE Form 278-T public filings (Office of Government Ethics).\n\nAsset Class Mix - Before and After (By Count)\n\nAlmost all bonds for 9 months. The May '26 filing - 98% equities.\n\nAug 2025 - Apr 2026 (14 filings, 1,726 transactions):\nMuni bonds 73%\nCorp bonds 14%\nBank preferred 11%\nOthers 2%\n\nAlmost entirely fixed income.\n\nMay 14, 2026 (1 filing, 3,642 transactions):\nEquities 98%\nOthers 2%\n\nAlmost entirely equities.\n\nAsset Class Mix - By Dollar Values\n\nSame story - May '26 predominantly equities. Muni bonds typically trade in larger lot sizes than equities, so dollar volume can run higher even with fewer transactions.\n\nAug 2025 - Apr 2026: ~$781M (63% of all volume)\nMuni 67%\nCorp 27%\nBank pref 2%\nOthers 4%\n\nMay 14, 2026: ~$454M (37% of all volume)\nEquities 84%\nOthers (mostly ETFs) 16%\n\nNote: Disclosure has range buckets. Values shown are midpoint sums - useful for general direction, not exact.\n\nMost Active Tickers\n\nTop 12 by total activity (buy + sell midpoint), Jan - Mar 2026.\n\n| Ticker | Company | Trades | Buy $M | Sell $M | Activity $M |\n|---|---|---:|---:|---:|---:|\n| MSFT | Microsoft Corporation | 22 | 5.25 | 16.89 | 22.14 |\n| AMZN | Amazon.com, Inc. | 21 | 5.42 | 16.32 | 21.75 |\n| META | Meta Platforms, Inc. | 20 | 1.67 | 16.57 | 18.24 |\n| VIG | Vanguard Div Apprec ETF | 1 | 0.00 | 15.00 | 15.00 |\n| VOO | Vanguard S&P 500 ETF | 5 | 6.75 | 6.00 | 12.75 |\n| NVDA | NVIDIA Corporation | 15 | 4.20 | 2.71 | 6.91 |\n| ORCL | Oracle Corporation | 17 | 3.62 | 3.26 | 6.88 |\n| CDW | CDW Corporation | 6 | 3.08 | 3.00 | 6.08 |\n| AAPL | Apple Inc. | 12 | 4.80 | 0.96 | 5.76 |\n| ACN | Accenture plc | 14 | 1.46 | 3.42 | 4.89 |\n| NFLX | Netflix, Inc. | 19 | 0.95 | 3.58 | 4.53 |\n| UBER | Uber Technologies | 16 | 3.39 | 0.97 | 4.36 |\n\nTop Buys\n\nLargest positions bought during Jan - Mar 2026 (gross BUY $M, midpoint).\nVOO Vanguard S&P 500 ETF - $6.75M\nAMZN Amazon.com, Inc. - $5.42M\nMSFT Microsoft Corporation - $5.25M\nAAPL Apple Inc. - $4.80M\nNVDA NVIDIA Corporation - $4.20M\nCOST Costco Wholesale - $4.05M\nIEMG iShares MSCI Emerging ETF - $3.75M\nMSI Motorola Solutions - $3.71M\nORCL Oracle Corporation - $3.62M\nADBE Adobe Inc. - $3.56M\n\nTop Sells\n\nLargest positions sold during Jan - Mar 2026 (gross SELL $M, midpoint).\nMSFT Microsoft Corporation - $16.89M\nMETA Meta Platforms - $16.57M\nAMZN Amazon.com, Inc. - $16.32M\nVIG Vanguard Div Apprec ETF - $15.00M\nVOO Vanguard S&P 500 ETF - $6.00M\nNFLX Netflix, Inc. - $3.58M\nACN Accenture plc - $3.42M\nORCL Oracle Corporation - $3.26M\nPLTR Palantir Technologies - $3.23M\nDIS The Walt Disney Company - $3.22M\n\nTrade-Action Timing Flagged by the Press (1 of 2)\n\nQ1 2026 OGE disclosures, alongside same-company actions by the administration.\n\nNVDA - NVIDIA Corporation. Bought $500K-$1M on Jan 6. A week later, on Jan 13, Commerce approved Nvidia H200 chip sales to China.\n\nAMD - Advanced Micro Devices. Bought $50K-$100K on Jan 6. On Jan 13, Commerce authorised AMD chip sales to China in the same rule. At least $740K in AMD across Q1.\n\nAXON - Axon Enterprise. Bought $1M-$5M on Feb 10. On Mar 2, ICE outlined a $220M five-year Taser contract - specifications only Axon products meet.\n\nDELL - Dell Technologies. Bought $1M-$5M on Feb 10. Dell family pledged $6.25B to Trump Accounts in Dec 2025. Trump endorsed Dell at the White House on May 8.\n\nTrade-Action Timing Flagged by the Press (2 of 2)\n\nPLTR - Palantir Technologies. Bought ~$247K-$630K across Q1 (seven March transactions, several unsolicited). In February, Palantir struck a billion-dollar DHS contract.\n\nBA - Boeing Company. Multiple Boeing buys during Q1 (a major US government contractor). On May 14 in Beijing, Trump announced Xi had agreed to a 200-plane order.\n\nINTC - Intel Corporation. Began buying INTC in early March (multiple unsolicited). Follows the US government's late-2025 decision to acquire a 9.9% stake in Intel.\n\nLLY - Eli Lilly and Company. Bought up to $680K across seven Q1 purchases (four unsolicited), first on Jan 6. CMS launched a Medicare GLP-1 pilot with a Jan 8 deadline.\n\nPress Sources\nCNBC, 14 May 2026 - China will order 200 Boeing jets, Trump tells Fox News\nCNBC, 14 Jan 2026 - Trump administration clears way for Nvidia H200 chip sales to China\nCNBC, 15 May 2026 - Trump touted Palantir on Truth Social after buying the company's stock, records show\nCNBC, 15 May 2026 - Trump went big on tech stocks in first quarter of 2026, new filings show\nWashington Post, 19 May 2026 - Trump discloses thousands of stock trades, some in companies he influences\nCNN, 22 Aug 2025 - The US takes a 10% stake in Intel as part of Trump's big tech push\nNOTUS, 14 May 2026 - Trump Bought Corporations' Stock as His Administration Boosted Their Business\nNOTUS, 2 Mar 2026 - ICE Plans to Spend $220 Million on Tasers\nKFF Health News, 18 May 2026 - Trump Bought Stock in Drugmaker as His Government Boosted Its Obesity Drugs\nYahoo Finance, 14 May 2026 - Trump revamps portfolio, adding Nvidia and other AI names\nTheStreet, 11 May 2026 - Trump praised a company at the White House and the stock soared\nTheStreet, 18 May 2026 - Trump brokerage bought Broadcom, Synopsys, Dell, Intel in Q1\n\nWhat the White House and Trump Organization Say\n\nVerbatim, as reported.\n\n\"President Trump's investment holdings are maintained exclusively through fully discretionary accounts independently managed by third-party financial institutions with sole and exclusive authority over all investment decisions. Trades are executed and portfolios are balanced through automated investment processes and systems administered by those institutions. Neither President Trump, his family, nor The Trump Organization plays any role in selecting, directing, or approving specific investments. They receive no advance notice of trading activity and provide no input regarding investment decisions or portfolio management of any kind.\"\nThe Trump Organization (spokesperson, statement carried by CNBC). CNBC, May 19 2026.\n\nDisclaimer and Notes\n\nImportant context for everything in this deck.\n\nNot financial advice. This deck is educational analysis of a public filing. Nothing here is a recommendation to buy, sell, or hold any security. Make your own decisions, talk to a registered advisor if you need one.\n\nPosition disclosure. I hold positions in some of the tickers mentioned in this deck.\n\nBest effort, no guarantees. OGE Form 278-T filings are PDFs with thousands of rows. The data was extracted using LLM vision models and document parsers, cross-checked across methods, with about 50 rows reviewed manually. Errors are possible. Treat the numbers as estimates, not audited figures.\n\nDollar values are midpoint sums. OGE discloses each trade as a range bucket, not an exact figure. Every dollar total here is the sum of the midpoints of those buckets. Useful for direction and scale. Not exact.\n\nUse at your own risk. I make no claim of accuracy, completeness, or fitness for any purpose. If you act on anything in this deck, that's on you, not me.\n\nMethodology. Refer to the 'Methodology' tab in the tool for detailed process and explanations.\n\nAmar Harolikar - Decision Sciences & Applied AI - tigzig.com - Apps for Analytics, Quants and Automation.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QRep - Security Reports, Quants Agent\n\nExplore: Quant apps hub, Analysis decks, APIs and MCP\n\nMore posts: Track Trump's Stock Trades. Public Filing Explorer. 3,568 Trades, 986 Tickers, $220M-$750M. Searchable on TREMOR., TREMOR. Six Free Tools for Banking and Insurance Analysts. Multi-Layer Validated. Built by a Practitioner., Three Tools I Built for Market Analysis - QSUITE, TREMOR, QREP, S&P 500 Drawdown Analysis with QREP"
},
{
"slug": "trump-trades-filings-aug2026",
"title": "What Trump Has Been Buying, Selling and Holding. The Filings, Extracted and Searchable.",
"tags": [
"portfolio-quants",
"database-ai"
],
"content": "What Trump Has Been Buying, Selling and Holding. The Filings, Extracted and Searchable.\n\nIf you want to see what Trump has been buying, selling and holding, the filings are on my site, extracted and searchable. Trades, assets, income, liabilities, gifts.\n\nTwo kinds of filing sit in there. The trade reports come in within 30 to 45 days of each trade and cover securities only, 15 filings and 5,368 trades so far.\n\nThe split is not even. For nine months it was mostly muni and corporate bonds, the earlier 14 filings have 1,726 trades between them.\n\nThen one filing on 14 May came in with 3,642 trades, almost all equities. That May one you can go through trade by trade. The earlier 14 are in there as per-filing summaries.\n\nAnd the annual disclosure, the 927 page one, which covers the whole year, 28,007 records including 21,285 trades.\n\nYou can use it two ways. Search and filter in the tool, roll the trades up by company, see them plotted against the S&P 500. Or download the lot as CSV or JSON and run your own cuts.\n\nFree, no sign-up.\n\ntigzig.com ➜ TREMOR ➜ Trump Trades.\n\nCompliance note\n\nNot investment advice. Educational analysis of public documents. Past Trump trades are no guide to future returns. Consult a qualified investment advisor. I hold positions in some of the tickers that show up in the data.\n\nOn how the data was pulled\n\nThe two filings were not the same job at all.\n\nThe trade reports arrive as scans. Seemed like someone ran the paper through an office copier and OCRs it, so you are working off a picture of a page. I ran several extraction methods over it and took the value the methods agreed on, with a fourth independent method redoing a sample of pages, and then a separate checker pass I kept away from. It found things. I fixed them.\n\nThe annual is the opposite. It came out as a proper digital document rather than a scan, so extraction is character exact. Two unrelated parsing engines, reconciled row by row, and I only kept the dataset because the two agreed on every field. After that, blind AI vision rereads on sampled pages, with any disagreement settled against a high resolution image of the printed page.\n\nAggregates were checked against independently reported figures as well, down to the exact dollar on the big income lines.\n\nAnd even after all the checking, an extraction error is still possible. I cannot rule it out. Every filing on the tool carries a link to the original OGE document, so if a number looks off, you can go and read it against the source.\n\nA few caveats\nValues on these forms are ranges, not exact numbers. That is how the form is built. So every securities total uses the midpoint of the range, and the open ended ones, over 50 million and so on, use the lower bound. Treat them as approximations.\nBusiness income is different, those are exact dollars as filed.\nThe annual does not repeat trades that were already filed on the trade reports. OGE rules allow that and this filing uses it. So neither document on its own is the full year. The two together are.\nFiling quirks are kept as filed. Odd capitalisation, the occasional typo, names cut short by the filer's own formatting. I did not tidy them, because then you would be reading my version rather than the document.\nTwo things are not in there. The 54 page trademark schedule, which is not structured data, and Part 4, which was filed as not applicable.\nDownloads are per table as CSV, or the whole filing as one JSON.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QRep - Security Reports, DATS-4 Database AI Suite\n\nExplore: Analysis hub, Tremor API docs, API and MCP catalog\n\nMore posts: Trump's 2025 annual financial disclosure: all 28,007 records, extracted and searchable, TREMOR Opened With Sixteen Charts in a Fixed Grid. That Was My View. Now You Can Change All of It., Shiller CAPE Is at 41.2. In 145 Years the Only Period Higher Was the Dot-Com Peak., Why Trump Is Going After Jamie Dimon Right Now. Two Fronts - Debanking + Crypto. The DOJ Subpoena Made It Real."
},
{
"slug": "trump-trades-reports-oct2026",
"title": "What Has Trump Been Buying and Selling? His Latest Trade Reports, Extracted and Downloadable",
"tags": [
"portfolio-quants",
"database-ai"
],
"content": "What Has Trump Been Buying and Selling? His Latest Trade Reports, Extracted and Downloadable\n\nWhat has President Donald Trump been buying and selling, and when? Trump's latest trade reports are on my site - extracted and downloadable. Use the interactive tool to run your own cuts, or download the data and analyze it offline.\n\nTwo types of reports - the 2025 annual disclosure, released by the Office of Government Ethics on 30 June 2026, and the periodic trade reports, the latest posted on 22 September 2026.\n\nThe trade reports have changed in size and in what they hold\n13 reports posted between August 2025 and April 2026, with trades from January 2025 to March 2026, had 1,657 trades between them, totalling $367 million to $1.15 billion, and 73% were municipal bonds.\nThe 6 reports posted since May, all with 2026 trades, have 8,443 trades totalling $581 million to $1.97 billion, and 95% are stocks. The four latest cover one month each, at 1,051 to 1,375 trades a month.\nAcross all 19 reports that is 10,100 trades, totalling $948 million to $3.12 billion.\n\nThe biggest trading days by dollar value (2026)\n10 February: Microsoft, Amazon and Meta sold, $5 million to $25 million each\n17 April: Microsoft, Amazon and Meta bought, each again in the $5 million to $25 million band\n20 July: Microsoft and Amazon sold, again $5 million to $25 million each\n7 April: bond sales of $22 million to $89 million, led by a US Treasury note at $5 million to $25 million\n\nThe 2025 annual disclosure\n\nThe 2025 annual disclosure lists investment-account holdings and cash, outside his businesses, of $858 million to $2.87 billion at the end of 2025, and 21,285 trades totalling $600 million to $1.86 billion. Trades already on a trade report are mostly left out, so the annual is largely the 2025 stock-account trades the public had not seen. Together the two cover his reported securities trades for 2025.\n\nA few data notes\nEvery value on these forms is a range, so each total above is the lower ends added up and the upper ends added up.\nThe trade reports are scans. Each page of the five largest reports, 8,374 trades, was read by several independent methods, with each value taken where the methods agreed.\nThe 14 smaller reports, 1,726 trades, were read once by an AI vision model.\nEvery filing on the tool links to the original OGE document, and the full data is a CSV download.\nThe trades are made in accounts that the White House says are managed by independent financial institutions.\n\nTool, data and downloads at:\nhttps://www.tigzig.com/tremor/tools/trump-trades\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QDesk - Quant Report Desk\n\nExplore: Analysis hub, Trump trades methodology, What is new\n\nMore posts: What Trump Has Been Buying, Selling and Holding. The Filings, Extracted and Searchable., Trump's 2025 annual financial disclosure: all 28,007 records, extracted and searchable, Track Trump's Stock Trades. Public Filing Explorer. 3,568 Trades, 986 Tickers, $220M-$750M. Searchable on TREMOR., Trump Trades. From Mostly Bonds to 3,642 Equity Trades in One Filing. Free Tool to Search Every Transaction."
},
{
"slug": "trump-vs-dimon-debanking-crypto-two-fronts-jun2026",
"title": "Why Trump Is Going After Jamie Dimon Right Now. Two Fronts - Debanking + Crypto. The DOJ Subpoena Made It Real.",
"tags": [
"portfolio-quants"
],
"content": "Why Trump Is Going After Jamie Dimon Right Now. Two Fronts - Debanking + Crypto. The DOJ Subpoena Made It Real.\n\nPublished: 2026-06-13\n\nWhy is Trump going after Jamie Dimon right now, five years after JPMorgan shut his accounts weeks after Jan 6, and why this particular week? The legal war front: Feb 2021 debanking, Aug 2025 Fair Banking order, Jan 2026 Trump sues Dimon for $5 billion, and for months the lawsuit just sat there - lawyers filing papers, nothing moving. This week it got teeth: on 10 Jun 2026 the DOJ (through Jeanine Pirro's office) subpoenaed JPMorgan, Bank of America and Wells Fargo over the debanking. Private lawsuits limping along are one thing; the federal machinery moving in behind them is a different beast. Running alongside is the crypto fight - Trump signed the GENIUS Act last summer (the big stablecoin law) and wants the US to be 'the crypto capital of the world'; his family has crypto ventures of their own. Dimon has been the loudest bank voice fighting the next bill (CLARITY Act) and told the Coinbase CEO to his face he is full of it. Then on 5 Jun 2026 - days before the subpoenas landed - the big banks announced they are building their own blockchain money, a shared tokenized deposit network, so customer deposits stay locked in the regulated banking system and do not bleed out into the stablecoins Trump has been backing. Did the crypto fight cause the DOJ to move? No paper trail. But the incidents line up too close to wave away - read the motive for yourself.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/trump-vs-dimon-debanking-crypto-two-fronts-jun2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/trump-vs-dimon-debanking-crypto-two-fronts-jun2026 with header Accept: text/markdown\n\nTags\nportfolio-quants\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: Analysis deck, Quant Apps, APIs and MCP\n\nMore posts: Dimon Was Right. More Cockroaches Arrive. UK Lender MFS Collapses - $2.5B Across 12+ Global Banks., Bank Loss Rates Look Calm. The Stuff Underneath - Consumer, Private Credit, Non-Banks - Says Otherwise., Banks Keep Funding the Non-Banks. $1.65T in Q1 2026, Up 42% in Five Quarters. US Banks NDFI Update."
},
{
"slug": "try-text-to-sql-on-real-data-gb-files-multi-million-rows",
"title": "Try Text-to-SQL on Real Data - Multi-Million Rows & GB+ Sizes",
"tags": [
"text-to-sql",
"database-ai",
"duckdb"
],
"content": "Try Text-to-SQL on Real Data - Multi-Million Rows & GB+ Sizes\n\nTwo Clicks. Zero Setup. No Database, No Server, No Login needed. With 9 LLM options, Python & SQL to torture your data till it confesses.\n\nApp live here: www.tigzig.com/analyzer\n\n## Related Post: Large File Upload for Database AI Text-to-SQL Apps: A Practical Guide\n\nWhat's New\n\nI've enhanced the sample datasets in my Database AI app (DATS-4). Previously, the test files were tiny, 50-60 rows. Now there's a full range: 64 rows to 11.8 million rows. File sizes from 14 KB to 1.6 GB.\n\nFor the 1.6 GB file, setup takes around 9 minutes. Fully automated: database creation, file upload, agent ready.\n\nThe Datasets\nRBI Cards & ATM Statistics: 14 KB, 64 rows. July 2025 data covering 60-70 banks.\nTour de France - Riders History: 974 KB, 10K rows. Race rankings from 1903 to 2025. Over 120 years of cycling history.\nIPL - Indian Premier League: 41 MB, 278K rows. Ball-by-ball data from 2003 to Sep 2025.\nODI - One Day International: 206 MB, 1.6 million rows. Ball-by-ball records, 2003 to Sep 2025.\nCricket Combined (ODI, T20, County, IPL): 697 MB, 5.2 million rows. 2003 to Sep 2025.\nCricket Extended (all formats including Test, T20 Blast): 1.6 GB, 11.8 million rows. 2003 to Sep 2025.\n\nTwo Clicks to Analytics-Ready\n\nSetup is two clicks\nGo to the app Tigzig Analyzer\nGo to Datasets. Pick one.\nSelect 'Use Temporary Database'\n\nThat's it. The app creates a temporary database, uploads your data, extracts the schema, and connects it to the AI agent. You're ready to query. For small files, setup takes 20-30 seconds. For the largest file, 2-3 minutes. Backend is neon.com - which provisions a Postgres database in less than a second via an API call.\n\nHow to Explore\n\nOnce setup completes, you're in the chat interface. Use the pre-built prompts. Each dataset has a sample prompt. Hit the copy icon, paste, run. These are structured queries: ranking systems, derived metrics, comparisons. The more specific the better. Avoid generic 'analyze this'. AI can't read your mind yet.\n\nOr explore data with:\n\"Show 5 sample rows in table format\"\n\"Have advanced analyst run EDA: univariates and categorical freqs, share results as table and charts\"\n\nCheck Agent Reasoning\n\nClick to see the SQL the agent generated. Useful for validation and learning.\n\nLLM Options\n\n9 models available for advanced analysis. Choose based on quality needs and cost tolerance.\n\n| Model | Type | Quality | Cost |\n|-------|------|---------|------|\n| Gemini 2.0 Flash | Best Value | 75 | Lowest |\n| Qwen3 Max | Good | 80 | Low |\n| Gemini 2.5 Flash | Good | 85 | Low |\n| KIMI K2 Thinking | High Variance | 85 | High |\n| Deepseek-R1-0158 | Great Quality | 90 | Med |\n| GPT-4.1 | Great Quality | 90 | Med |\n| Gemini 3 Pro | Good | 95 | High |\n| GPT-5.1 | Top Quality | 100 | High |\n| Claude 4.5 Sonnet | Topmost Quality | 115 | High |\n\nFor detailed cost and quality comparisons based on live testing, see Gemini 3 Pro Added to Database AI Suite. Tested Against Claude Sonnet 4.5 and GPT-5.1\n\nSummary: Claude still leads. GPT-5.1 is solid. Gemini 3 Pro lands third.\n\nWhat Else Can the App Do\n\nThe sample dataset feature is just one entry point. DATS-4 is a full database AI suite. Here's what's available:\n\nDatabase Connections\nConnect to any remote Postgres or MySQL database with your own credentials\nOr use the on-the-fly temporary database for quick tests\nPaste credentials in any format (URI, table, plain text). AI parses it.\n\nTwo Agents\nGeneral Analyst: fast execution for direct queries, data pulls, standard charts. Powered by GPT-4.1-mini.\nAdvanced Analyst: multi-step reasoning for complex analysis. Choice of 9 LLMs for the reasoning step. Execution by GPT-4.1.\n\nFile Uploads\nUpload CSV or tab-delimited files directly\nUpload to temporary database or your own database\nAI-powered schema detection. You don't define columns. It figures it out.\n\nWorking Tables & Export\nCreate derived tables, run transformations, merge datasets\nExport any table to CSV or pipe-delimited file\nDownload for offline analysis in Excel or other tools\n\nTable Viewer\nInteractive data grid for all uploaded files\nFilter, sort, drill down to record level\nOn-the-fly descriptive statistics and data quality metrics\n\nPDF Output\nAgent can convert analysis output to PDF (text only, charts not yet supported)\nStructure and content customizable via natural language instructions\n\nPython Charts & Stats\nIntegrated Python sandbox (e2b Code Interpreter)\nGenerate charts: bar, line, scatter, heatmap, violin, radar, box plots\nRun statistical analysis: Chi-square, ANOVA, correlation matrices, distributions\n\nLogs\nDetailed logging of API calls and agent actions\nFirst line of debugging for when things go wrong\n\nTechnical Note on File Uploads\nDownload: App downloads compressed file from GitHub repo.\nCompression (frontend): Uncompressed CSV or TXT uploads are compressed using the browser CompressionStream API without loading full file into memory.\nTemporary database provisioning: A temporary Postgres database is created via Neon with automatic role setup and unique credentials.\nFile upload to backend: Compressed file is sent to the FastAPI SQL connector.\nMemory efficient file handling: Backend streams file to disk in 32MB chunks to prevent RAM bloat.\nDecompression: Backend decompresses .gz files when needed, streaming to disk in 32MB chunks.\nAI powered schema detection: Backend samples first 5 lines, detects delimiter, and sends data to OpenAI for schema inference.\nTable creation: Empty table is created using the detected schema.\nSmart upload path selection: Postgres uses in memory COPY for uncompressed files under 100MB and streamed COPY from temp file for larger or compressed files. MySQL always streams in 100K row batches using Polars or Pandas with executemany inserts.\nAgent handoff: After upload, schema plus credentials and sample rows are handed to the Database Agent.\nConfirmation: App confirms environment readiness and the Agent confirms schema receipt.\n\nOpen Source\n\nAll open source. Docs and source code accessible from the app (hit Docs in top nav). Guides and posts at tigzig.com. The app has 7 major components, each with its own GitHub repo:\nMain App (React UI)\nFastAPI Server: Database Connector\nFastAPI Server: Neon DB Creation\nFlowise Agent Schemas\nProxy Server\nMCP Server: Markdown to PDF\nQuant Agent Backend\n\nFull build guide and architecture docs available in the Docs section.\n\nVisual Guide: GB File Upload Process\n\n10 slides explaining the upload architecture and data flow.\n\nLinks\nApp: www.tigzig.com/analyzer\nLLM Cost & Quality Assessment: Gemini 3 Pro Test Results\nField Guide (PDF): DATS-4 Database AI Suite\nGuides & Posts: tigzig.com\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals are in the PDF and slide images on the HTML page.\n\nAnalyze Your Data In Natural Language (Text-to-SQL)\n\nWant to Test ?\n\nSample Data: 11.8 Million Rows. 1.6 GB\n\n2 Click Setup\n\nNo Server – Database – Login needed\n\nFree to Try\n\nwww.tigzig.com/analyzer\n\nStep 1: Select Dataset\n\nfrom 14KB to 1.6GB\n\nFrom 64 rows to 11.8 M Rows\n\nStep 2: Select Temp. DB\n\nStep 3 : App Confirms Setup\n\nBackend is neon.com — which provisions a Postgres database in less than a second via an API call.\n\nStep 4: Request in regular English\n\nUse the prompt provided or explore on your own\n\nFor complex queries ask AI to send request to Advanced Analyst\n\nStep 5: Agent Shares Analysis\n\nStep 6 : View Agent Reasoning\n\nStep 7: Iterate\n\nIterate\n\nValidate\n\nCreate Working Tables\n\nMerge\n\nSummarize\n\nExport Table\n\nCreate PDF\n\nFull Post\n\nhttps://www.tigzig.com/post/try-text-to-sql-on-real-data-gb-files-multi-million-rows\n\nDatasets: How to Explore\n\nLLM Options: Costing & Pricing\n\nTechnical Note on File Uploads\n\nWhat Else Can the App Do\n\nDatabase Connections\n\nTwo Agents\n\nPython Charts and Stats\n\nFile Uploads & Table Exports\n\nPDF Output\n\nLogs\n\nOpen Source / Links\n\nCricket data from Cricsheet, published under the Open Data Commons Attribution License 1.0 (ODC-BY).\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, BRIQ, Cricsheet File Processor\n\nExplore: Database AI Apps hub, Database Query Tool (MCP), Browse all apps\n\nMore posts: Cricket (ODI/T20) & Tour de France stats from a Custom GPT connected to 3 live databases., Two models added to Database AI Suite this week: GPT-5.1 and KIMI 2 Thinking., Instant Database Setup for AI Apps. With Neon.com, BRIQ App: DuckDB AI in Browser - 500MB Files, 4M+ Records, No Database Setup"
},
{
"slug": "two-models-added-to-database-ai-suite-this-week-gpt-5-1-and-kimi-2-thinking",
"title": "Two models added to Database AI Suite this week: GPT-5.1 and KIMI 2 Thinking.",
"tags": [
"database-ai",
"text-to-sql"
],
"content": "Two models added to Database AI Suite this week: GPT-5.1 and KIMI 2 Thinking\n\nKIMI 2 Thinking is close to Gemini 2.5 Flash in quality. Costs are higher and more volatile. GPT-5.1 is 20% cheaper than GPT-5 for database execution. Token bloat is down. Default reasoning set to none instead of medium.\n\nFor advanced analysis planning (single API call, reasoning step)\nGemini 2.5 Flash: strong quality, predictable costs\nKIMI 2: similar quality, costs are higher and more volatile\nGPT-5.1: high quality, high cost\nClaude Sonnet 4.5: highest quality, highest cost\nGPT-5 / Gemini-2.5-Pro: avoid on APIs - major token and cost bloat\n\nFor database execution (where 80% of cost sits)\nGPT-4.1-mini: my current workhorse for simple to medium work, very cost effective\nGPT-4.1: complex queries, strong at debugging, 30% cheaper than GPT-5.1\nGPT-4o-mini: high volume work, half the cost of GPT-4.1-mini\n\nGPT-5-mini still has token bloat. Default reasoning is medium. For regular work, GPT-4.1-mini is better. For more firepower ramp up stepwise → GPT-4.1 → GPT-5.1\n\nPlanning vs execution costs\n\nIn multi-step advanced analysis workflow, for a single iteration, planner runs once. Execution agent runs 7 to 10 queries and debugs. Planning is 20% of cost. Execution is 80%.\n\nAvoid multi-step unless needed\n\nMulti-step workflows multiply costs fast. Single step: approx. $0.40 per 100 questions with a GPT-4.1-mini. Advanced analysis: approx. $15 per 100 questions depending on LLM. Use multi-step multi-agent workflows only when needed.\n\nDATS-4 (Database AI Suite v4)\n\nDatabase AI app. Connects to Postgres or MySQL. Two workflows: simple text-to-sql and advanced multi-step analysis. Supports python charts, table upload, export, PDF reports.\n\nTry it\n\nSample button uploads test data to temporary Postgres instance. Use sample prompts. Or upload your files to a temp DB / connect to your DB.\n\nNote\n\nPublic app routes through my backend. Sandbox testing only. For production, deploy on your servers.\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, BRIQ, ChatGPT to Any Database\n\nExplore: Database AI Apps, Database Query Tool (MCP), APIs and MCP\n\nMore posts: Gemini 3 Pro on Database AI Suite - Claude 4.5 still leads, Sonnet 4.5. Released yesterday. Now live on DATS-4 SQL Agent Suite. Solid upgrade, but more 4.2 than 4.5., LLM Costing for Database AI Apps. Live Experience. Live App. Open Source"
},
{
"slug": "two-of-the-best-resources-i-ve-seen-on-building-agentic-ai-one-from-manus-one-from-anthropic",
"title": "Two of the best resources I've seen on building agentic AI. One from Manus, one from Anthropic.",
"tags": [
"ai-coders"
],
"content": "Two of the best resources I've seen on building agentic AI. One from Manus, one from Anthropic.\n\nRequired reading. They are dense, practical guides from builders who have clearly felt the pain of agents going rogue and burning credits.\n\nMulti-Agent Research System (Anthropic)\n\nMulti-Agent Research System\n\nContext Engineering for AI Agents (Manus)\n\nContext Engineering for AI Agents\n\nMy own work offers a field guide for a specific use case: using agents for live databases (TEXT2SQL/ Natural Language-to-SQL). The full field guide and open-source apps are on my site: www.tigzig.com\n\n\nRelated\n\nTools: BRIQ, Database AI Apps, Quants Agent\n\nExplore: Guides hub, API/MCP hub, Full app catalog\n\nMore posts: Claude Code Was Getting Dumber. Stopped It from Auto-Spawning Sub-Agents. Performance Came Back Instantly., Mistakes I Made Building Text-to-SQL Agents in Live Projects. My 2025 Learnings, Anthropic Released Claude Managed Agents Yesterday, Live on TREMOR Today"
},
{
"slug": "upstash-redis-vercel-rate-limiting",
"title": "Upstash Redis on Vercel - The Tool I Didn't Know I Needed",
"tags": [
"infrastructure",
"security",
"open-source"
],
"content": "Upstash Redis on Vercel - The Tool I Didn't Know I Needed\n\nIt's like a Python dictionary that lives on a server, remembers things across requests, and cleans itself up. Useful for rate limiting, caching, and many other stuff.\n\nsrc: /images/blog/upstash-redis-explainer.png\nalt: Upstash Redis architecture and key-value storage explained\n\nI come from data science - two years into building full-stack AI tools, my apps kept getting hammered. Bots, scrapers, someone just deciding to hit an endpoint a thousand times. The serverless functions on Vercel were taking the hits directly - and if any of those functions talked to a paid API, that's money going out while I'm asleep.\n\nRate limiting was the obvious fix. But there's a problem with serverless functions - they forget everything after each request. Every time one runs, it starts fresh. No memory of who called it before, how many times, or when. So how do you tell it \"this IP has already made 5 requests in the last 15 minutes, block the 6th\"?\n\nYou need something external to hold that thing...that's what led me to Redis / Upstash.\n\nWhat is Redis, and how is it different from a database?\n\nRedis is a key-value store. Not a database with tables and columns - just keys and values. Think of it like a Python dictionary that lives on a server, in memory, and responds in under a millisecond.\n\nThe simplest way to understand it is to just look at what's actually stored:\n\n``\nKEY VALUE\n───────────────────────────────── ─────\nlogin_attempts:192.168.1.45 3\nlogin_attempts:10.0.0.22 1\nfeedback:global:2026-03-10 14\nvigil:query:172.16.0.5 87\n`\n\nThat's it. That's the whole structure. Each row is one entry - a name (the key) and a value. No tables, no columns, no schemas. The key is a string. The value can be a string, a number, a list, or a set. Everything lives in RAM, not on disk. That's why it's fast - reads take under a millisecond vs 50-200ms for a typical Postgres query.\n\nThere's one important thing that makes Redis genuinely different: you can set a TTL (time-to-live) on any key. Means \"remember this for X seconds, then auto-delete.\" So when someone from IP 192.168.1.45 hits your login endpoint, you create a counter for that IP and tell Redis to forget it after 15 minutes. Every attempt increments the counter. After 5 attempts - blocked. When 15 minutes are up, the key vanishes on its own. No cleanup code, no cron jobs. Redis handles it.\n\nThat's the thing that makes it very useful for rate limiting.\n\nWhy not just use Postgres for this?\n\nMy first instinct was to create a rate_limits table in my existing Neon Postgres. Three reasons I didn't:\n\nLatency. Postgres takes 50-200ms including connection setup. Redis over HTTP takes 1-5ms. Rate limit check runs on every single request before any other logic. That difference adds up fast across multiple apps.\n\nAuto-expiry. In Postgres, I'd need a cleanup job - \"delete all rate limit rows older than 15 minutes.\" In Redis, I set a TTL and it disappears by itself.\n\nOverhead. Another connection string, another schema, another thing to monitor. Upstash on Vercel is literally two environment variables. No migrations, no connection pooling config.\n\nWhat is Upstash?\n\nTraditional Redis requires a running server - install it, configure it, keep it alive. If the whole point of using Vercel is to avoid managing servers, running your own Redis defeats the purpose.\n\nUpstash is Redis-as-a-service built for serverless. Speaks HTTP (REST API), so it works from any serverless function without persistent connections. On Vercel, it's a first-class integration - click a button in the dashboard, it provisions a Redis instance and injects two environment variables: KVRESTAPIURL and KVRESTAPITOKEN. That's it, you're connected.\n\nFor small apps and indie projects, a managed service like Upstash makes sense. You don't want to manage infrastructure for a rate limiter. Large-scale setups with dedicated DevOps teams might prefer their own Redis cluster for control and cost at volume. But for building and shipping tools fast, Upstash is to Redis what Vercel is to hosting.\n\nMany apps, one database\n\nI have some 10-12 apps using the same single Upstash instance. Did not create 12 databases. Created one.\n\nRedis is a flat namespace - no tables, no schemas. The trick is key naming. Each app uses a prefix:\n\n`\ntremor:login:192.168.1.45 → 3\nvigil:query:10.0.0.12 → 87\nfeedback:global:2026-03-10 → 14\nqrep:yf:172.16.0.5 → 22\n`\n\nSame database, same env vars, keys never collide because of the prefix. Like separate folders inside one filing cabinet. The free tier allows up to 10 databases, but for rate limiting there's no need to split - just use prefixes. Keys self-organize and auto-expire. The database essentially cleans itself.\n\nFree Tier - What You Get\n500,000 commands/month - roughly 16,000/day. Each rate limit check is 2-3 commands. Across 10-12 apps with moderate traffic, I've never come close.\n256 MB storage - way more than needed when keys auto-expire. My entire rate limiting dataset across all apps is probably a few kilobytes at any given moment.\n1 free database - up to 10 total, additional ones at $0.50/month each.\n\nHow it works for my rate limiters\n\nA note before the implementation: rate limiting needs to happen at multiple layers\n\nThis section covers rate limiting at the serverless function layer on Vercel. But depending on your app architecture, you need it in more places - at the Cloudflare edge, at your FastAPI backend, and here at the serverless layer. Each layer uses a different IP source and a different tool. And the IP itself is a trap - the wrong header gives you a spoofable value, not the real client IP.\n\nI go deep into all of that in a separate post: Are You Rate Limiting the Wrong IPs?\n\nThis section is specifically about the serverless layer - where Upstash Redis fits in.\n\nWhen a request comes in:\nExtract the client IP (Vercel gives you this in its own headers - the real TCP/IP connection IP, not the spoofable X-Forwarded-For header)\nBuild a key like app:endpoint:192.168.1.1`\nIncrement the counter in Redis\nSet TTL so it auto-expires\nIf count exceeds threshold, return 429\n\nWhat this looks like across my apps:\nBug report endpoint - 5 per 15 minutes per IP, global cap of 50/day across all users\nAI feedback endpoint - 10 per hour per IP\nLogin pages - 5 attempts per 15 minutes. Fail past that, IP gets blocked in Redis until admin unblocks\nDuckDB query proxies - 200 requests per 60 seconds per IP\nYahoo Finance proxies - 60 per minute per IP\nImage uploads - 15 per hour per IP\n\nEvery implementation uses fail-open. If Redis is down, the request goes through. You don't want your entire app to break because the rate limiter is temporarily unavailable.\n\nWhat Else Can You Do With It\n\nI'm only using it for rate limiting right now. But there's a huge amount more it can do...I haven't used these, but here are some common uses I can think of:\n\nAPI response caching. Run a heavy SQL query or hit an expensive API. Cache the result in Redis with a 5-minute TTL. Next request gets the cached result instantly instead of waiting 3 seconds.\n\nSession storage. Store user sessions without a database table. Key is the session token, value is the user object. TTL handles expiry. No session table to manage.\n\nFeature flags. Roll out a new feature to 10% of users. Store the flag in Redis. Change it instantly without redeployment.\n\nReal-time leaderboards. Redis sorted sets maintain rankings automatically. Add a score, Redis keeps the order. Could be useful for analytics dashboards showing top-performing funds or stocks.\n\nSliding window analytics. Track events in a time window - \"how many API calls in the last 60 seconds\" with sub-second precision. Same pattern as rate limiting, but pointed at usage dashboards and monitoring.\n\nConfiguration store. Store app config that changes often - API endpoints, threshold values, toggle switches. Read from Redis, update without redeployment.\n\nThe Bigger Picture\n\nThe pattern is always the same: set a key, check a key, let it expire. Whether rate limiting, caching, or tracking - same three operations. Once you get that, you start seeing where it fits.\n\n500,000 monthly commands, 256 MB storage, no server to manage, sub-millisecond response times - on the free tier. For the kind of tools I build, that's more than enough.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, VIGIL - India Red Flag Events, QRep - Security Reports\n\nExplore: Browse all apps, APIs and MCP, Database AI apps\n\nMore posts: Are You Rate Limiting the Wrong IPs? A SlowAPI Story., Per-IP rate limit on Cloudflare free plan - the gotchas, Your API Key Is Visible in the Browser. Even if you put it as Vercel's 'secret' backend env variable., 2026 Infra Guide for AI Tool Builders - Part 5: Perimeter Security on Cloudflare (Free Plan), Instant Database Setup for AI Apps. With Neon.com"
},
{
"slug": "us-bank-loan-data-q2-2026-tremor-sep2026",
"title": "US Bank Loan Data Is Now Updated Through Q2 2026 on TREMOR. NY Fed, FDIC and the FFIEC Call Reports, All Three in One Place.",
"tags": [
"portfolio-quants",
"database-ai"
],
"content": "US Bank Loan Data Is Now Updated Through Q2 2026 on TREMOR. NY Fed, FDIC and the FFIEC Call Reports, All Three in One Place.\n\nUS bank loan data is now updated on Tremor. Free to download. NY Fed, FDIC and the FFIEC Call Reports, all three now through Q2 2026.\n\nThree sets in one place. Every US bank added up, back to 2001. 754 banks one at a time, with their capital and their lending to non-bank financial firms. And households, from the NY Fed and Equifax credit panel, back to 1999.\n\nQuick points\nI was expecting this quarter to look worse, and most of it came off the top instead.\nOn the bank books, the FDIC side, loans 90 days late or on nonaccrual are 0.93 percent of all US bank lending. Cards came down from 1.64 to 1.44 percent. Commercial and real estate eased a little as well.\nOn the household side, the NY Fed panel, which counts balances already written off, the headline 90 plus flow came down from 3.26 percent two quarters ago to 2.57 percent.\nTwo went the other way. Auto is the only retail number that rose on the bank side, and on the NY Fed side mortgage 90 plus flow has now risen in six of the last seven quarters. Levels on most of these are still well above 2019, so one quarter of easing is not much to go on.\n\nFree, no signup, no API key.\nTREMOR tool and manual download: tigzig.com/tremor ➜ US Credit / US NDFI\nTREMOR API hub: tigzig.com/apis/tremor\n\nThe bank and NY Fed numbers will not tally\n\nIf you are comparing the bank and NY Fed numbers side by side, they won't tally ... they are measuring different things.\nBank FDIC and the Call Reports come from what lenders file about their own books.\nThe NY Fed panel is built the other way round, from credit files, a 5 percent random sample of Equifax records, so it picks up consumer borrowing wherever it sits, including the part that has moved off bank balance sheets.\nThe definitions differ as well. NY Fed 90 plus includes balances that have already been charged off and are still showing on the credit report. FDIC pulls charge-offs out of delinquency and reports them separately.\nSo credit cards at 12.92 percent on the NY Fed panel and credit cards at 1.44 percent on the FDIC books are both right. The lens differs.\nThe bank charge-off number has two more wrinkles. It is a flow for the period, a quarter or a year, where the NY Fed figure is a stock of balances sitting on the credit file, and a written-off balance stays on that file for years after the bank has taken it off its own books.\nThe bank number is also net of recoveries, and collections keep arriving on loans written off, which is minused from the charge-off line in that quarter.\n\ntigzig.com/tremor\n\nGetting the data itself\n\nOn getting the data itself, all of it can be downloaded by hand from the tool. The FDIC and NY Fed sets are also on the open API, no key and no signup, so you can pull them straight into a sheet or a script. Details are on the Tremor page, or at the API hub: tigzig.com/apis/tremor\n\nThe bank by bank file is the one exception. That is still a manual download from the tool.\n\nMy earlier writeups on Q2 bank loans and delinquencies\nFDIC, and what the loan book is now made of: FDIC Q2 2026 numbers are live on TREMOR\nNY Fed households, auto and mortgage against Q3 2007: NY Fed Q2 2026 household debt\nBank lending to non-banks, 754 banks one at a time: US banks had $2.8 trillion committed to non-bank lenders\n\nUS bank loan data through Q2 2026, now live on TREMOR"
},
{
"slug": "us-bank-ndfi-interactive-analytics-tool",
"title": "US Bank Exposure to Non-Bank Lenders - Now in an Interactive Analytics Tool",
"tags": [
"portfolio-quants"
],
"content": "US Bank Exposure to Non-Bank Lenders - Now in an Interactive Analytics Tool\n\nPublished: 2026-04-04\n\nInteractive analytics tool for US bank exposure to non-bank lenders. 629 banks, $1.57 trillion in NDFI loans, 5 quarters of FFIEC Call Report data. CET1 concentration, ACL reserve coverage, bank-level drill downs, data explorer with 55+ columns, methodology with MDRM codes, 7 levels of validation. Full dataset download.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/us-bank-ndfi-interactive-analytics-tool\nMarkdown of the HTML page: send GET https://tigzig.com/post/us-bank-ndfi-interactive-analytics-tool with header Accept: text/markdown\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nUS Banks & Non-Bank Lending\n\nHow Deep is the Exposure?\n\nPrivate credit stress is rising. BDC defaults climbing, redemptions spiking, JPMorgan already marking down loans. Through mechanisms like back leverage, banks have $1.57 trillion on these lenders. If private credit losses start cascading, where does the stress land? And is there a systemic risk?\n\nFFIEC Call Reports Analysis- Q424 to Q42526th March 2026 | TigZig.com\n\nAmar Harolikar, ACA\n\nDecision Sciences & Applied AI\n\nHow much stress can the system absorb?\n\nAt 5-10% loss rates on non-bank lending of $1.5T , the banking system absorbs the shock - existing loan loss reserves ($190B) cover it entirely. However, that would exhaust the reserves, leaving little coverage for rest of the $13T of loans.\n\nConcentration risk is a further concern - 8 Large Financial Institutions (LFIs, Assets>$100B) have NDFI exposure exceeding 100% of their CET1 capital and 14 LFIs have ACL coverage of less than 15%. A 10% loss rate exhausts existing loan loss reserves at these banks. At 15-20%, losses start eroding CET1 capital directly - for the most concentrated banks, this could breach regulatory minimums and trigger forced capital raises\n\nCorrelated losses: The bigger risk isn't NDFI in isolation - it's NDFI losses hitting alongside private credit stress, rising consumer delinquencies, commercial real estate distress, and macro headwinds. Simultaneous stress across multiple asset classes is what turns manageable stress into systemic risk\n\nThe US Banking and NDFI Universe\n\n629 NDFI lending banks cover nearly 88% of banks industry assets\n\nNDFI Loans as of Q425 were >$1.5T and represent 13.7% loan book of these 629 banks\n\nThis single portfolio now exceeds 80% of NDFI lending banks’ CET1 Capital\n\nAmar Harolikar. Decision Sciences & Applied AI | Data: FFIEC Call Reports | tremor.tigzig.com -> US-NDFI\n\nStress scenarios\n\nAt 5-10% loss rates, loan loss reserves ($190B) absorb the entire hit. No capital erosion.\n\nAt 15%+, losses start eating into CET1 capital directly. The 'After ACL' column shows how much passes through.\n\nAt 30%, $280B hits capital - that's 14.4% of system CET1. Manageable at system level, but not at the 41 banks where NDFI already exceeds 100% of their capital.\n\nFor reference: US bank writedowns in 2008 were ~$885B (IMF GFSR, April 2010 )\n\nAmar Harolikar. Decision Sciences & Applied AI | Data: FFIEC Call Reports | tremor.tigzig.com -> US-NDFI\n\n41 banks have NDFI / CET1 ratio exceeding 100%\n\nAdd the 75-100% bucket, and we have 59 banks holding 80% of all NDFI dollars with exposure above 75% of CET1 (Common Equity – Tier 1)\n\nThis is across all 629 NDFI lending banks Q425 (representing 88% of total banking assets)\n\nAmar Harolikar. Decision Sciences & Applied AI | Data: FFIEC Call Reports | tremor.tigzig.com -> US-NDFI\n\nTop banks by NDFI to CET1 concentration\n\nLarge Financial Institutions (Assets >$100B) ranked by NDFI concentration relative to CET1 capital. 8 of the top 15 have NDFI exceeding 100% of CET1\n\nHigh NDFI/CET1 does not imply capital inadequacy - these banks maintain Tier 1 ratios well above regulatory minimums.\n\nThe ratio highlights concentration: how much of a bank's core capital is exposed to a single fast-growing loan category\n\nAmar Harolikar. Decision Sciences & Applied AI | Data: FFIEC Call Reports | tremor.tigzig.com -> US-NDFI\n\n62 banks have ACL coverage of less than 10%\n\nFor these 62 banks holding 59% of NDFI dollars a 10% loss on NDFI portfolio will wipe out 100% of their ACL (Allowance for Credit Losses) reserves\n\nAmar Harolikar. Decision Sciences & Applied AI | Data: FFIEC Call Reports | tremor.tigzig.com -> US-NDFI\n\nACL coverage of NDFI balances\n\nLarge Financial Institutions (Assets >$100B) ranked by ACL (Allowance for Credit Losses). For 8 of top 15, a 5% loss will wipe out the bank level ACL provisions\n\nLow ACL/NDFI at institutional banks reflects a historical bet: corporate and institutional lending has had very low loss rates.\n\nIf private credit stress breaks that pattern, these banks have no reserve cushion. Losses go straight to capital.\n\nAmar Harolikar. Decision Sciences & Applied AI | Data: FFIEC Call Reports | tremor.tigzig.com -> US-NDFI\n\nWhat are NDFI Loans\n\nNDFI stands for Nondepository Financial Institution - any financial company that lends money but doesn't take deposits like a bank. Think private credit funds, mortgage companies, fintech lenders, PE funds.\n\nBanks lend heavily to these entities. Starting December 2024, regulators required banks to break out these loans into five categories:\n\nMortgage lenders - companies that originate, service, or securitize home and commercial mortgages\n\nBusiness lenders - direct lenders, private debt funds, BDCs, CLOs, finance companies. This is the bulk of private credit\n\nPrivate equity funds - banks provide capital call lines and NAV loans to PE funds\n\nConsumer lenders - fintech and non-bank companies lending directly to consumers\n\nOther NDFIs - broker-dealers, insurance companies, hedge funds, pension funds, and other financial entities\n\nAmar Harolikar. Decision Sciences & Applied AI | Data: FFIEC Call Reports | tremor.tigzig.com -> US-NDFI\n\nWhat's the issue with NDFI loans? Back leverage\n\nAfter 2008, regulators restricted banks from making risky loans directly - especially in real estate. Banks didn't stop. They found a workaround: lend to non-bank financial companies (NDFIs) who then make the risky loans instead. Same risk, different label on the balance sheet. Also called back leverage\n\nExample - commercial real estate (CRE):\n\nA bank can't easily make a high-risk CRE loan anymore\n\nInstead, it lends to a mortgage company or private credit fund (an NDFI)\n\nThat fund makes the risky CRE loan using the bank's money\n\nOn the bank's books this shows up as a ‘loan to a financial institution’ - not CRE. The risk is identical, the classification is different\n\nHow big is this? As of Q4 2025:\n\nNDFI balances are ~$1.6T , making up more than 13% of NDFI lenders’ loan book\n\nAmar Harolikar. Decision Sciences & Applied AI | Data: FFIEC Call Reports | tremor.tigzig.com -> US-NDFI\n\nNDFI grew by 35% over past year\n\nAmar Harolikar. Decision Sciences & Applied AI | Data: FFIEC Call Reports | tremor.tigzig.com -> US-NDFI\n\nTop 12 banks account for 77% of all NDFI loans\n\nAmar Harolikar. Decision Sciences & Applied AI | Data: FFIEC Call Reports | tremor.tigzig.com -> US-NDFI\n\nBanks have committed another 65% to NDFI\n\nCommitments are additional lines granted to NDFI’s but not yet utilized\n\nAs of Q425, banks had committed additional $1T over and above the $1.5T outstanding\n\nIf fully utilized that takes potential exposure to >$2.6T\n\nAmar Harolikar. Decision Sciences & Applied AI | Data: FFIEC Call Reports | tremor.tigzig.com -> US-NDFI\n\nNDFI Nonaccrual rate is currently just 0.20%\n\nNonaccrual means the bank has stopped recognizing interest income on the loan, signaling it no longer expects full collection.\n\nNDFI nonaccrual rate is 0.20%, well below the bank-wide rate of 0.67%. However these losses can balloon suddenly as downstream delinquencies start and NDFI bankruptcies start to come in.\n\nAlready starting: JPMorgan has marked down the value of private credit loans held as collateral and reduced borrowing capacity for private credit firms - a preemptive move targeting software sector loans where AI disruption fears are driving a downcycle. (CNBC, March 11 2026)\n\nAmar Harolikar. Decision Sciences & Applied AI | Data: FFIEC Call Reports | tremor.tigzig.com -> US-NDFI\n\nNonaccrual rate computed for banks with $10B+ in total assets (the only banks required to report NDFI subcategories). These banks hold 98% of total NDFI by value. Rate covers three subcategories that report nonaccrual: Mortgage Credit, Business Credit, and PE Funds (70% of NDFI). Consumer Credit and Other NDFIs (remaining 30%) do not report nonaccrual separately. The true combined NDFI nonaccrual rate could be higher or lower depending on stress in these unreported segments.\n\nWhere is the stress coming from?\n\nPrivate credit market showing cracks. Blackstone's flagship fund posted its first loss in 3 years. Blue Owl halted redemptions. BDCs trading at 0.73x NAV, a 27% implied discount on their loan books.\n\nMorgan Stanley projects direct lending defaults hitting 8%, approaching COVID peaks. Half of direct lending borrowers had negative free operating cash flow (IMF, October 2025).\n\nNDFI loans are the transmission channel. Banks don't hold the risky loans directly. They lend to NDFIs who do. When downstream borrowers default, the stress travels back to bank balance sheets through this $1.6T pipeline.\n\nDetailed analysis of private credit stress, market signals, and macro backdrop in companion deck : Private Credit - The $2.7 Trillion Shadow Lending Market Is Showing Cracks\n\nAmar Harolikar. Decision Sciences & Applied AI | Data: FFIEC Call Reports | tremor.tigzig.com -> US-NDFI\n\nFurther Reading\n\nS&P Global: US banks' NDFI lending pace reaccelerates in Q4 2025 (Feb 2026) - industry aggregate tallies at $1,569B, consistent with my FFIEC analysis\n\nMoody's: US banks' private credit loan exposure nears $300 billion (Oct 2025) - breakdown by NDFI subcategory\n\nFDIC: FDIC Quarterly Banking Profile Q4 2025 PDF - for validating industry numbers\n\nGeorge Smith Partners: The Loans Banks Aren't Supposed to Hold Anymore by William Letzer (March 2026)\n\nFederal Reserve Board of Governors, FEDS Notes: Bank Lending to Private Credit: Size, Characteristics, and Financial Stability Implications (May 23, 2025)\n\nFederal Reserve Bank of Boston: Current Policy Perspectives 25-8, May 2025\n\nABA Banking Journal: Loans to non-depository financial institutions: new granularity and a rapidly growing segment (Feb 2026)\n\nMayer Brown: Private Credit Portfolio Back Leverage (March 2025)\n\nPYMNTS: Private Credit's Gold Rush Draws Regulators and Bank Warnings (March 2026)\n\nCNBC: JPMorgan reins in lending to private credit firms, marks down software loans (March 2026)\n\nAmar Harolikar. Decision Sciences & Applied AI | Data: FFIEC Call Reports | tremor.tigzig.com -> US-NDFI\n\nUS-NDFI Tool Live with data from Q424 to Q425\n\ntremor.tigzig.com ➜ US-NDFI\n\nCoverage\n\nBalances\n\nCommitments\n\nCredit Quality\n\nExposure\n\nBank View\n\nData Explorer\n\nMethodology\n\nValidations\n\nInteractive tool with drill down, bank view and full data download\n\nAmar Harolikar. Decision Sciences & Applied AI | Data: FFIEC Call Reports | tremor.tigzig.com -> US-NDFI\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Private Credit and NDFI Tracker, Market tools hub, Analysis hub\n\nMore posts: Banks Keep Funding the Non-Banks. $1.65T in Q1 2026, Up 42% in Five Quarters. US Banks NDFI Update., NDFI Analytics - US Bank Exposure to Non-Bank Lenders, US Banks & Non-Bank Lending - How Deep is the Exposure? Is there a Systemic Risk?, Private Credit - The $2.7 Trillion Shadow Lending Market Is Showing Cracks"
},
{
"slug": "us-banks-ndfc-lending-systemic-risk-2026",
"title": "US Banks & Non-Bank Lending - How Deep is the Exposure? Is there a Systemic Risk?",
"tags": [
"portfolio-quants"
],
"content": "US Banks & Non-Bank Lending - How Deep is the Exposure? Is there a Systemic Risk?\n\nPublished: 2026-03-26\n\nUS banks hold $1.57 trillion in loans to non-bank lenders facing redemptions and defaults. 41 banks have this single portfolio exceeding 100% of core capital. 62 banks have loan loss reserves covering less than 10% of exposure. Analysis from FFIEC Call Reports for 629 banks across 5 quarters, cross-validated against FDIC, FRED and S&P Global.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/us-banks-ndfc-lending-systemic-risk-2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/us-banks-ndfc-lending-systemic-risk-2026 with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/TREMORNDFCEXPOSURE.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nUS Banks & Non-Bank Lending\n\nHow Deep is the Exposure?\n\nPrivate credit stress is rising. BDC defaults climbing, redemptions spiking, JPMorgan already marking down loans. Through mechanisms like back leverage, banks have $1.57 trillion on these lenders. If private credit losses start cascading, where does the stress land? And is there a systemic risk?\n\nFFIEC Call Reports Analysis- Q424 to Q42526th March 2026 | TigZig.com\n\nAmar Harolikar, ACA\n\nDecision Sciences & Applied AI\n\nHow much stress can the system absorb?\n\nAt 5-10% loss rates on non-bank lending of $1.5T , the banking system absorbs the shock - existing loan loss reserves ($190B) cover it entirely. However, that would exhaust the reserves, leaving little coverage for rest of the $13T of loans.\n\nConcentration risk is a further concern - 8 Large Financial Institutions (LFIs, Assets>$100B) have NDFI exposure exceeding 100% of their CET1 capital and 14 LFIs have ACL coverage of less than 15%. A 10% loss rate exhausts existing loan loss reserves at these banks. At 15-20%, losses start eroding CET1 capital directly - for the most concentrated banks, this could breach regulatory minimums and trigger forced capital raises\n\nCorrelated losses: The bigger risk isn't NDFI in isolation - it's NDFI losses hitting alongside private credit stress, rising consumer delinquencies, commercial real estate distress, and macro headwinds. Simultaneous stress across multiple asset classes is what turns manageable stress into systemic risk\n\nThe US Banking and NDFI Universe\n\n629 NDFI lending banks cover nearly 88% of banks industry assets\n\nNDFI Loans as of Q425 were >$1.5T and represent 13.7% loan book of these 629 banks\n\nThis single portfolio now exceeds 80% of NDFI lending banks’ CET1 Capital\n\nStress scenarios\n\nAt 5-10% loss rates, loan loss reserves ($190B) absorb the entire hit. No capital erosion.\n\nAt 15%+, losses start eating into CET1 capital directly. The 'After ACL' column shows how much passes through.\n\nAt 30%, $280B hits capital - that's 14.4% of system CET1. Manageable at system level, but not at the 41 banks where NDFI already exceeds 100% of their capital.\n\nFor reference: US bank writedowns in 2008 were ~$885B (IMF GFSR, April 2010 )\n\n41 banks have NDFI / CET1 ratio exceeding 100%\n\nAdd the 75-100% bucket, and we have 59 banks holding 80% of all NDFI dollars with exposure above 75% of CET1 (Common Equity – Tier 1)\n\nThis is across all 629 NDFI lending banks Q425 (representing 88% of total banking assets)\n\nTop banks by NDFI to CET1 concentration\n\nLarge Financial Institutions (Assets >$100B) ranked by NDFI concentration relative to CET1 capital. 8 of the top 15 have NDFI exceeding 100% of CET1\n\nHigh NDFI/CET1 does not imply capital inadequacy - these banks maintain Tier 1 ratios well above regulatory minimums.\n\nThe ratio highlights concentration: how much of a bank's core capital is exposed to a single fast-growing loan category\n\n62 banks have ACL coverage of less than 10%\n\nFor these 62 banks holding 59% of NDFI dollars a 10% loss on NDFI portfolio will wipe out 100% of their ACL (Allowance for Credit Losses) reserves\n\nACL coverage of NDFI balances\n\nLarge Financial Institutions (Assets >$100B) ranked by ACL (Allowance for Credit Losses). For 8 of top 15, a 5% loss will wipe out the bank level ACL provisions\n\nLow ACL/NDFI at institutional banks reflects a historical bet: corporate and institutional lending has had very low loss rates.\n\nIf private credit stress breaks that pattern, these banks have no reserve cushion. Losses go straight to capital.\n\nWhat are NDFI Loans\n\nNDFI stands for Nondepository Financial Institution - any financial company that lends money but doesn't take deposits like a bank. Think private credit funds, mortgage companies, fintech lenders, PE funds.\n\nBanks lend heavily to these entities. Starting December 2024, regulators required banks to break out these loans into five categories:\n\nMortgage lenders - companies that originate, service, or securitize home and commercial mortgages\n\nBusiness lenders - direct lenders, private debt funds, BDCs, CLOs, finance companies. This is the bulk of private credit\n\nPrivate equity funds - banks provide capital call lines and NAV loans to PE funds\n\nConsumer lenders - fintech and non-bank companies lending directly to consumers\n\nOther NDFIs - broker-dealers, insurance companies, hedge funds, pension funds, and other financial entities\n\nWhat's the issue with NDFI loans? Back leverage\n\nAfter 2008, regulators restricted banks from making risky loans directly - especially in real estate. Banks didn't stop. They found a workaround: lend to non-bank financial companies (NDFIs) who then make the risky loans instead. Same risk, different label on the balance sheet. Also called back leverage\n\nExample - commercial real estate (CRE):\n\nA bank can't easily make a high-risk CRE loan anymore\n\nInstead, it lends to a mortgage company or private credit fund (an NDFI)\n\nThat fund makes the risky CRE loan using the bank's money\n\nOn the bank's books this shows up as a ‘loan to a financial institution’ - not CRE. The risk is identical, the classification is different\n\nHow big is this? As of Q4 2025:\n\nNDFI balances are ~$1.6T , making up more than 13% of NDFI lenders’ loan book\n\nNDFI grew by 35% over past year\n\nTop 12 banks account for 77% of all NDFI loans\n\nBanks have committed another 65% to NDFI\n\nCommitments are additional lines granted to NDFI’s but not yet utilized\n\nAs of Q425, banks had committed additional $1T over and above the $1.5T outstanding\n\nIf fully utilized that takes potential exposure to >$2.6T\n\nNDFI Nonaccrual rate is currently just 0.20%\n\nNonaccrual means the bank has stopped recognizing interest income on the loan, signaling it no longer expects full collection.\n\nNDFI nonaccrual rate is 0.20%, well below the bank-wide rate of 0.67%. However these losses can balloon suddenly as downstream delinquencies start and NDFI bankruptcies start to come in.\n\nAlready starting: JPMorgan has marked down the value of private credit loans held as collateral and reduced borrowing capacity for private credit firms - a preemptive move targeting software sector loans where AI disruption fears are driving a downcycle. (CNBC, March 11 2026)\n\nNonaccrual rate computed for banks with $10B+ in total assets (the only banks required to report NDFI subcategories). These banks hold 98% of total NDFI by value. Rate covers three subcategories that report nonaccrual: Mortgage Credit, Business Credit, and PE Funds (70% of NDFI). Consumer Credit and Other NDFIs (remaining 30%) do not report nonaccrual separately. The true combined NDFI nonaccrual rate could be higher or lower depending on stress in these unreported segments.\n\nWhere is the stress coming from?\n\nPrivate credit market showing cracks. Blackstone's flagship fund posted its first loss in 3 years. Blue Owl halted redemptions. BDCs trading at 0.73x NAV, a 27% implied discount on their loan books.\n\nMorgan Stanley projects direct lending defaults hitting 8%, approaching COVID peaks. Half of direct lending borrowers had negative free operating cash flow (IMF, October 2025).\n\nNDFI loans are the transmission channel. Banks don't hold the risky loans directly. They lend to NDFIs who do. When downstream borrowers default, the stress travels back to bank balance sheets through this $1.6T pipeline.\n\nDetailed analysis of private credit stress, market signals, and macro backdrop in companion deck : Private Credit - The $2.7 Trillion Shadow Lending Market Is Showing Cracks\n\nFurther Reading\n\nS&P Global: US banks' NDFI lending pace reaccelerates in Q4 2025 (Feb 2026) - industry aggregate tallies at $1,569B, consistent with my FFIEC analysis\n\nMoody's: US banks' private credit loan exposure nears $300 billion (Oct 2025) - breakdown by NDFI subcategory\n\nFDIC: FDIC Quarterly Banking Profile Q4 2025 PDF - for validating industry numbers\n\nGeorge Smith Partners: The Loans Banks Aren't Supposed to Hold Anymore by William Letzer (March 2026)\n\nFederal Reserve Board of Governors, FEDS Notes: Bank Lending to Private Credit: Size, Characteristics, and Financial Stability Implications (May 23, 2025)\n\nFederal Reserve Bank of Boston: Current Policy Perspectives 25-8, May 2025\n\nABA Banking Journal: Loans to non-depository financial institutions: new granularity and a rapidly growing segment (Feb 2026)\n\nMayer Brown: Private Credit Portfolio Back Leverage (March 2025)\n\nPYMNTS: Private Credit's Gold Rush Draws Regulators and Bank Warnings (March 2026)\n\nCNBC: JPMorgan reins in lending to private credit firms, marks down software loans (March 2026)\n\nUS-NDFI Tool Live with data from Q424 to Q425\n\ntremor.tigzig.com ➜ US-NDFI\n\nCoverage\n\nBalances\n\nCommitments\n\nCredit Quality\n\nExposure\n\nBank View\n\nData Explorer\n\nMethodology\n\nValidations\n\nInteractive tool with drill down, bank view and full data download\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Private Credit & NDFI Tracker, Quant Apps hub, Analysis hub\n\nMore posts: Banks Keep Funding the Non-Banks. $1.65T in Q1 2026, Up 42% in Five Quarters. US Banks NDFI Update., NDFI Analytics - US Bank Exposure to Non-Bank Lenders, The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top., Private Credit - The $2.7 Trillion Shadow Lending Market Is Showing Cracks, Bank Loss Rates Look Calm. The Stuff Underneath - Consumer, Private Credit, Non-Banks - Says Otherwise."
},
{
"slug": "us-banks-ndfi-q1-2026-update-may2026",
"title": "Banks Keep Funding the Non-Banks. $1.65T in Q1 2026, Up 42% in Five Quarters. US Banks NDFI Update.",
"tags": [
"portfolio-quants"
],
"content": "Banks Keep Funding the Non-Banks. $1.65T in Q1 2026, Up 42% in Five Quarters. US Banks NDFI Update.\n\nPublished: 2026-05-30\n\nQ1 2026 update on US bank lending to non-depository financial institutions (NDFI) - the channel funding private credit. Balances hit $1.65T, up 42% in five quarters and now 14.2% of the entire bank loan book (up from 11.1%) or 7.1% of total assets (up from 5.6%). 622 banks. Top 10 hold 73% of all NDFI loans. Subcategory mix: Business Credit $435B (26%), PE Funds $383B (23%), Mortgage Credit $368B (22%), Other NDFIs $269B (16%), Consumer Credit $166B (10%). Top lenders: JPMorgan $238B, Wells Fargo $228B, Bank of America $213B, Goldman $123B, Citi $123B. Add undrawn commitments of $1.06T and total exposure hits $2.7T. The Fed, FSB, OCC, IMF, ECB, BoE and DOJ all flagging it; Fitch private-credit default rate at a record 6.0%; Bank of England calling it a 'market for lemons'. Tool live at tigzig.com/tools/us-ndfi (TREMOR) with bank-level balances, commitments, credit quality and exposure - free, no login, sourced directly from quarterly FFIEC Call Reports (Schedule RC-C). Continues the four-dispatch private-credit series (Mar 24, Mar 26, May 8, May 28).\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/us-banks-ndfi-q1-2026-update-may2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/us-banks-ndfi-q1-2026-update-may2026 with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/USBANKSNDFI_MAY2026.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nTIGZIG · US Banks · NDFI · Q1 2026 Update\n\nUS bank loans to non-depository financial institutions hit $1.65 trillion in Q1 2026, up 42% in five quarters, now 14.2% of the entire bank loan book. This is the channel that funds private credit.\n\nCentral banks and regulators are all now flagging it: the Fed, FSB, OCC, IMF, ECB, BoE and the DOJ.\n\nQ1 2026 - Now updated. Explore the data yourself, free, no login: tigzig.com → TREMOR → US NDFI. Balances · Commitments · Credit Quality · Exposure.\n\n622 Banks. $1.65T. Five Quarters Straight Up.\n\nThe headline numbers (Q1 2026):\nNDFI loans: $1,650B.\n5-quarter growth: +42% (+5.1% this quarter).\n% of bank assets: 7.1% (up from 5.6%).\n% of bank loan book: 14.2% (up from 11.1%).\n\nThe quarter-by-quarter climb:\n\n| Quarter | NDFI Loans ($B) |\n|---|---:|\n| Q4 '24 | 1,159 |\n| Q1 '25 | 1,273 |\n| Q2 '25 | 1,383 |\n| Q3 '25 | 1,463 |\n| Q4 '25 | 1,569 |\n| Q1 '26 | 1,650 |\n\nFive quarters straight up. Not one quarter down.\n\nTop 10 banks hold 73% of all NDFI loans.\n\nWhat the loans fund (by NDFI subcategory, $B of balance)\nBusiness Credit - $435B (26%).\nPE Funds - $383B (23%).\nMortgage Credit - $368B (22%).\nOther NDFIs - $269B (16%).\nConsumer Credit - $166B (10%).\n\nWho holds it (top NDFI lenders, $B and share)\nJPMorgan Chase - $238B (14.4%).\nWells Fargo - $228B (13.8%).\nBank of America - $213B (12.9%).\nGoldman Sachs - $123B (7.5%).\nCitibank - $123B (7.5%).\n\nThe Bigger Picture - The Drawn Balance is Only Half the Story\n\nAdd the undrawn lines and exposure hits $2.7T. $1.65T already lent, plus $1.06T in commitments still to be drawn.\n\nAnd the stress is showing. The private-credit default rate just hit a record 6.0% (Fitch), though roughly 9 in 10 of those are stressed restructurings, not bankruptcies. The Bank of England calls it a \"market for lemons.\"\n\nThe previous four dispatches\n\nRead them all at tigzig.com/analysis.\n04 - May 28 - Is Private Credit Turning Into a Lemon? (latest)\n03 - May 8 - Red Flag in US Life Insurance: $2.4T in Private Credit\n02 - Mar 26 - US Banks & Non-Bank Lending: How Deep Is the Exposure?\n01 - Mar 24 - The $2.7T Shadow Lending Market Is Showing Cracks\n\nSource: \"Loans to nondepository financial institutions,\" reported by US banks in their quarterly FFIEC Call Reports (Schedule RC-C), Q1 2026. US-domiciled exposure only; foreign-branch lending not included. Served via Tremor. Default-rate and regulator references from the Tigzig private-credit series.\n\nAmar Harolikar - Decision Sciences & Applied AI - tigzig.com - Tools for analytics, quants and macro signals. Built by an analyst, for analysts.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: Private Credit and NDFI Tracker, Analysis hub, Quant apps hub\n\nMore posts: US Banks & Non-Bank Lending - How Deep is the Exposure? Is there a Systemic Risk?, NDFI Analytics - US Bank Exposure to Non-Bank Lenders, Private Credit - The $2.7 Trillion Shadow Lending Market Is Showing Cracks, Private Credit Default Rate Hits a Record 6.0% (2026): A Market for Lemons? Banks and Insurers Deep In It., Bank Loss Rates Look Calm. The Stuff Underneath - Consumer, Private Credit, Non-Banks - Says Otherwise."
},
{
"slug": "us-banks-nonbank-lending-q2-2026",
"title": "US Banks Had $2.8 Trillion Committed to Non-Bank Lenders at the End of Q2 2026, and $1.7 Trillion of It Was Already Drawn.",
"tags": [
"portfolio-quants",
"database-ai"
],
"content": "US Banks Had $2.8 Trillion Committed to Non-Bank Lenders at the End of Q2 2026, and $1.7 Trillion of It Was Already Drawn.\n\nUS banks had 2.8 trillion dollars committed to non-bank lenders at the end of Q2 2026, and 1.7 trillion of that was already drawn, totalling 14.3% of their loan book. Fitch reports a private credit default rate of 6.1% for the twelve months to July 2026, which its own release describes as remaining at a record high.\n\nThe drawn amount is 14.3 percent of these banks' own loans, up from 11.1 percent seven quarters ago, and it has gone up in every single quarter since banks first had to report it separately in Q4 2024.\n\nI have put the Q2 2026 numbers up on Tremor, bank by bank. 754 banks named across the seven quarters, 633 of them reporting in the latest one, with drawn balances, undrawn commitments, the five lending categories, capital ratios, and the nonaccrual and past due amounts. 37 fields per bank per quarter.\n\nSource is the FFIEC Call Reports.\n\nYou can use the tool in the browser or download the whole thing as a CSV, both free.\n\ntigzig.com/tremor ➜ US NDFI\n\nA small part of the bank population, and most of the money\n\nThe 633 banks reporting non-bank lending in Q2 2026 are roughly one in seven of all US banks by count, there are about 4,300 filing Call Reports, but between them they hold 88.7 percent of all US banking assets and 85.5 percent of all US bank loans.\n\nSo the banks lending to non-banks are a small part of the bank population and most of the money.\n\nThat also puts the 1.7 trillion in a wider frame. Measured against every US bank loan rather than only these banks, lending to non-banks is 12.2 percent. Measured against the loans of the banks doing the lending, it is 14.3 percent. Both are right, they are different denominators, and I use the second one in the tool because those are the banks with the exposure.\n\nA few notes on the data, for anyone who downloads it\n\nThe five lending categories, mortgage, business credit, private equity funds, consumer credit and other, are reported only by banks with 10 billion dollars or more in assets. Smaller banks file the non-bank total without the split, which is why about 32 billion of the 1.7 trillion sits unallocated when you add the categories up.\n\nThe commitment figure is the unused portion, so drawn plus undrawn gives you the total committed and there is no double counting between the two.\n\nCapital fields come out blank for banks that elected the Community Bank Leverage Ratio framework, because those banks do not file risk based ratios. Blank there means not reported, it does not mean zero, so do not divide by it.\n\nOn the FDIC cross-check, total assets agree on all 4,320 bank quarter records and the non-bank balance agrees on 4,319. The exception is M&T Bank in Q4 2024, where the FDIC shows about 1.4 billion more than the filing I read. I have not worked out which of the two is right.\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals are in the PDF and slide images on the HTML page.\n\nSlide 1\n\ntigzig.com\n ·\n the analyst's tool shed\n ·\n AI-agent first\n\n Bank-level lending to non-banks, now live on TREMOR\n\n tigzig.com\n → TREMOR → US NDFI\n (download available)\n\n FFIEC Call ReportsQ2 2026\n\n $2.8 trillion committed to non-banks\n\n $1.70tn\n Already lent\n\n $1.09tn\n Committed, not yet lent\n\n 14.3%\n of their loans\n\n Fitch's US private credit default rate hit 6.1% in the twelve months to\n July 2026, and has been at record highs since April.\n\nWhat is inside\n\n 1\n\nLending up 47% since Q4 '24\n\n 2\n\nThe biggest lenders are not the most concentrated\n\n 3\n\nA few banks have lent several times their own capital\n\n 4\n\nThe loss does not arrive as a delinquency\n\n 5\n\nThe regulators, and my earlier analysis\n\nWhat is in the data\n754 banks named, 633 in the latest quarter\n7 quarters, Q4 2024 to Q2 2026\n37 fields per bank per quarter\nDrawn balances and undrawn commitments\nFive lending categories\nCapital ratios\nNonaccrual and past-due amounts\n\n Amar Harolikar, ACADecision Sciences & Applied AI\n\n 1\n\nSlide 2\n\ntigzig.com\n ·\n the analyst's tool shed\n ·\n AI-agent first\n\n The balancesQ4 2024 to Q2 2026\n\n Seven quarters, and it has not paused\n\nUp 47% since Q4 2024\n\nLending to non-banks has gone from $1,159 billion to $1,704 billion\n since Q4 2024, and it has risen in every single quarter.\n\nAgainst these banks' own loan books it has gone from 11.1% to 14.3%.\n Roughly one dollar in seven that they lend now goes to another financial firm.\n\nQ4 2024 is the start because that is the first quarter banks had to report it\n separately. Before that it sat inside other loan lines.\n\nWhat it is lent against\n\nBusiness credit is the largest single category at $446 billion. This is\n the line that holds bank lending to private credit funds.\n\nAdd private equity funds at $410 billion and half the book is financing\n the two things most people mean when they say private credit.\n\nThe $32 billion not broken out is mostly banks under $10 billion in assets,\n which file the total and not the split.\n\n Amar Harolikar\n\n 2\n\nSlide 3\n\ntigzig.com\n ·\n the analyst's tool shed\n ·\n AI-agent first\n\n FFIEC Call ReportsQ2 2026\n\n Which banks lend most to non-banks\n\n Dollars lent on the left, and what share of that bank's own loan book it represents\n on the right. The two say different things: the biggest lenders are not the most\n concentrated.\n\nRead the right-hand column with that in mind. Goldman\n Sachs, Morgan Stanley and State Street are not primarily lenders. Loans are a third\n or less of their balance sheets and lean towards funds, so a high share of a small\n loan book is close to what you would expect. The $133 billion and the $83\n billion are real, and so is the exposure.\n\n Amar Harolikar\n\n 3\n\nSlide 4\n\ntigzig.com\n ·\n the analyst's tool shed\n ·\n AI-agent first\n\n FFIEC Call ReportsQ2 2026\n\n Now ranked by capital at risk\n\n Banks that have lent $1 billion or more to non-banks, 66 of them, ranked by that\n lending as a share of their own core capital. The top 20 are shown.\n\nRead it as loss absorption. A 20% loss on this book would take\n 16% of JPMorgan's core capital, 45% of Goldman Sachs' and more than all of\n Northpointe's. It does not say how concentrated a bank is: at Goldman and Morgan\n Stanley the loan book is a third or less of the balance sheet, at Northpointe and\n Axos it is almost the whole bank.\n\nAt system level it is contained. A 10% loss across the whole\n $1.70 trillion uses 8.6% of combined core capital. One small bank failing is contained.\n Several failing together is not. The Call Report names no counterparties, and\n nothing else published does either, so the overlap between them cannot be seen. What the FSB set out in May 2026 is\n the shape of it: layers of leverage stacked on one another, and a web of links\n between banks, funds and the same borrowers.\n\n Amar Harolikar\n\n 4\n\nSlide 5\n\ntigzig.com\n ·\n the analyst's tool shed\n ·\n AI-agent first\n\n The loss does not arrive as a delinquency\n\n 1\n\nBanks left the lending. They never left the financing.\n\nAfter 2008, regulators pushed banks out of risky middle-market lending. They\n came back as the financiers behind it: credit lines, warehouse facilities,\n leverage. The risk did not leave the banking system. It moved one step away\n from it.\n\n 2\n\nThe past-due buckets will not warn you.\n\nThese borrowers keep paying until they cannot, and many switch to paying in\n kind before they stop, which keeps a deteriorating loan looking current. When it\n does break, it usually breaks as a bankruptcy filing.\n\n 3\n\nAnd the loss never shows up as a charge-off.\n\nMost private credit losses are carried at fair value. They hit the profit and loss\n account directly and never pass through the delinquency or charge-off lines that\n everyone watches.\n\n Fitch Ratingsthe number that moves when the bank data does not\n\n 6.1%\n\n US private credit default rate, twelve months to July 2026.\n At record highs since April, across about 1,500 US private credit issuers. Published\n monthly.\n Fitch Ratings, 13 August 2026\n\n 5.2%\n\n MCO · Model-based Credit Opinion\n\n Fitch's own opinions on issuers, which feed the ratings on pooled\n assets such as middle-market CLOs. A record, up from 4.9%.\n\n 8.6%\n\n PMR · Privately Monitored Rating\n\n Ratings insurers need on the loans they hold, for regulatory\n capital. Down from 9.4%, a third straight monthly fall from a record 10.0% in March.\n\nAnd look at what counts as a default. Half the default events\n were an interest deferral or payment in kind replacing cash interest. Another 38% were\n maturity extensions under stress. Only 8% were an uncured missed payment.\n\n Amar Harolikar\n\n 5\n\nSlide 6\n\ntigzig.com\n ·\n the analyst's tool shed\n ·\n AI-agent first\n\n Go deeperwhat the regulators are saying\n\n The major regulators have written on this\n\nFinancial stability bodies on both sides of the Atlantic, and the\n international bodies, have published on private credit and on bank lending to\n non-banks in the past eighteen months. They\n do not agree on how serious it is.\nReport on Vulnerabilities in Private Credit Financial Stability Board · 6 May 2026\nFinancial Stability Report Federal Reserve · May 2026\nFinancial Stability Review European Central Bank · May 2026\nGlobal Financial Stability Report IMF · April 2026\nSemiannual Risk Perspective OCC · Spring 2026\nThis time is different? Speech by Sarah Breeden Bank of England · April 2026\nBank lending to private credit: size, characteristics and financial stability implications Federal Reserve, FEDS Note · 23 May 2025\nCould the growth of private credit pose a risk to financial system stability? Boston Fed, CPP 25-8 · 21 May 2025\n\n Amar Harolikar\n\n 6\n\nSlide 7\n\ntigzig.com\n ·\n the analyst's tool shed\n ·\n AI-agent first\n\n My earlier analysis on this\n\nPublished on tigzig.com\nBank loss rates look calm. The stuff underneath says otherwise 4 June 2026\nBanks keep funding the non-banks. $1.65 trillion in Q1 2026, up 42% in five quarters 30 May 2026\nPrivate credit default rate hits a record 6.0%: a market for lemons 28 May 2026\nRed flag in US life insurance. $2.4 trillion in private credit and offshore reinsurance 8 May 2026\nThe convergence risk: credit stress already past 2007 pre-crisis 29 April 2026\nNDFI Analytics: US bank exposure to non-bank lenders 31 March 2026\nUS banks and non-bank lending: how deep is the exposure, and is there a systemic risk 26 March 2026\nPrivate credit: the $2.7 trillion shadow lending market is showing cracks 24 March 2026\n\nDetailed sections, these and more\n\ntigzig.com\n → TREMOR → US NDFI\n the bank-level data behind this deck\n\ntigzig.com\n → Private Credit\n every analysis and live tool on this topic, in one place\n\n Amar Harolikar\n\n 7\n\nSlide 8\n\ntigzig.com\n ·\n the analyst's tool shed\n ·\n AI-agent first\n\n My read\n\nWhere it stands today\n\nI am not calling this a crisis. Nonaccrual on this book is 0.13% and it fell\n again this quarter, though page 5 is exactly why I would not lean on that. Roughly one dollar in seven that these banks lend now goes to\n another financial firm, and that share has risen in every quarter since it became\n separately reportable.\n\nWhat the money is lent against\n\nA large part of private credit sits with software companies and the businesses\n building out AI, the part of the market with a live valuation question over it.\n In Fitch's own data, technology software is among the largest sectors in private\n credit and has the lowest default rate of any of the big ones, 1.2%.\n\nThe consumer is already turning\n\nFitch's consumer products and services default rate went from 5.9% in July 2025\n to 9.9% in July 2026. In the NY Fed and Equifax consumer credit panel, US auto\n loans 90 days past due sit at 5.5%, a level beaten\n only once in the sixty-six quarters since 2010, and credit cards at 12.9% have been\n higher in only six of them. Both eased a little this quarter. That 90-day measure\n includes balances already charged off, repossessed or foreclosed, which the NY\n Fed groups as severely derogatory.\n\nWhy it could spread\n\nSo it is finely balanced. If AI valuations correct, or the economy weakens enough\n that insolvencies pick up, the losses land first on the private credit funds and\n then on the banks financing them. That is where contagion risk sits, because the\n market is layered: a bank lends to a fund, the fund lends to a company, and there is\n leverage at every level. Most of the regulators on page 6 make some version of\n that point.\n\n Amar Harolikar\n\n 8\n\nSlide 9\n\ntigzig.com\n ·\n the analyst's tool shed\n ·\n AI-agent first\n\n Sources\n\nThe bank data\n\nFFIEC Call Reports, Q2 2026, 633 banks reporting lending to non-banks.\n Loans to nondepository financial institutions became a separate reporting line in\n Q4 2024, which is why the series starts there.\n\nFDIC BankFind API for the cross-check. Every bank in every quarter was\n matched on RSSD ID and compared. Total assets agree on all 4,320 bank-quarter\n records. The NDFI balance agrees on 4,319, the exception being M&T Bank in\n Q4 2024, where FDIC shows $1.4 billion more than the filing we read.\n\nEverything else\n\nFitch Ratings for the private credit default rates, from the July 2026\n release published on 13 August 2026.\n\nNY Fed Household Debt and Credit Report, Q2 2026, for the auto and credit\n card figures.\n\nThe regulator and central bank sources are listed on page 6, and my own earlier\n analysis on page 7.\n\ntigzig.com\n\nBuilt by an analyst, for analysts. Track macros, compare securities, query\n databases. In your browser, in Excel, in Claude, or through an API.\n\n Amar Harolikar\n\n 9\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, DATS-4 Database AI Suite\n\nExplore: Market and macro tools, Private credit hub, API and MCP catalog\n\nMore posts: FDIC Q2 2026 Numbers Are Live on TREMOR. 102 Quarters of Bank Balances, Delinquency and Charge-Offs., NY Fed Q2 2026 Household Debt. The Headline Delinquency Flow Fell, but Auto and Mortgage Went the Other Way., Shiller CAPE Is at 41.2. In 145 Years the Only Period Higher Was the Dot-Com Peak., Reading Macro Charts on TREMOR? Send Them to the AI Agent Sitting There and Get a Read With the Latest News."
},
{
"slug": "us-consumer-lending-crisis-without-recession-may2026",
"title": "US Consumer Lending Is Already at Crisis-Level Numbers - Without a Recession. NY Fed Q1 2026 Data.",
"tags": [
"portfolio-quants"
],
"content": "US Consumer Lending Is Already at Crisis-Level Numbers - Without a Recession. NY Fed Q1 2026 Data.\n\nPublished: 2026-05-15\n\nNY Fed Q1 2026 release. Auto delinquencies past GFC peak. Cards near it. Mortgage stock calm but flow rate cracking. HELOC usage and delinquencies rising. Home prices down 9% from 2022 peak. U-6 unemployment back to pre-GFC 2007 levels. Underwater mortgages up in 45 states YoY. Crisis numbers without a recession.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/us-consumer-lending-crisis-without-recession-may2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/us-consumer-lending-crisis-without-recession-may2026 with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/USCONSUMERLENDINGCRISISMAY2026.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nUS Lending Crisis - II\n\nAmar Harolikar\n\nDecision Sciences & Applied AISpecialist – Banking Analytics\n\n14th May 2026\n\nTigzig.com\n\nQ1 2026 NY Fed data\n\nAuto past GFC peak. Cards near GFC peak. Mortgage cracking at the flow. And this is happening across the whole consumer landscape - without a recession.\n\ntremor.tigzig.com\n\nAuto 90+ DPD at 5.6%. Past the 5.3% GFC peak.NY Fed 90+ DPD includes accumulated severely derogatory balances (charge-offs, repos, foreclosures on file up to 7 years). Closer to a loss-rate measure than the bank-portfolio 90+ DPD definition.\n\nNY FED CONSUMER - ALL DEBT\n\nAggregate 90+ DPD at 3.4%. Already past the 3.1% pre-GFC Q3-2007 level. Steady climb over the last six quarters.\n\nSource: NY Fed Consumer Credit Panel / Equifax via Tigzig Tremor platform\n\nNote: NY Fed 90+ DPD is broader than the bank-portfolio definition - it includes accumulated severely derogatory balances (charge-offs, repos, foreclosures on file up to 7 years). Closer to a loss-rate measure.\n\nNY FED CONSUMER - SEGMENT MIX\n\nStress is broad-based. Cards and auto are loudest. But mortgage at 1.1% is rising fast off a pandemic-distorted base - more on that ahead.\n\nSource: NY Fed Consumer Credit Panel / Equifax via Tigzig Tremor platform\n\nNote: NY Fed 90+ DPD is broader than the bank-portfolio definition - it includes accumulated severely derogatory balances (charge-offs, repos, foreclosures on file up to 7 years). Closer to a loss-rate measure.\n\nNY FED CONSUMER - CREDIT CARDS\n\nCards at 13.1%, near 13.7% GFC peak. Different from 2008-09 - this time it has climbed without a recession.\n\nSource: NY Fed Consumer Credit Panel / Equifax via Tigzig Tremor platform\n\nNote: NY Fed 90+ DPD is broader than the bank-portfolio definition - it includes accumulated severely derogatory balances (charge-offs, repos, foreclosures on file up to 7 years). Closer to a loss-rate measure.\n\nNY FED CONSUMER - AUTO LOANS\n\nAuto at 5.6%. Past the 5.3% GFC peak. The first major consumer segment to break beyond GFC levels.\n\nSource: NY Fed Consumer Credit Panel / Equifax via Tigzig Tremor platform\n\nNote: NY Fed 90+ DPD is broader than the bank-portfolio definition - it includes accumulated severely derogatory balances (charge-offs, repos, foreclosures on file up to 7 years). Closer to a loss-rate measure.\n\nNY FED CONSUMER - MORTGAGE (STOCK)\n\nMortgage 90+ DPD at 1.1%. Half the 2.2% pre-GFC level - on the surface, looks fine. But the pace of the rise is the story. Tripled from 0.37% in Q3-2022. Even adjusting for the CARES Act forbearance distorting that trough, the climb off pre-pandemic norms is real. 2006-07 also started calm. The flow rate next is what to watch..\n\nSource: NY Fed Consumer Credit Panel / Equifax via Tigzig Tremor platform\n\nNY FED CONSUMER - MORTGAGE (FLOW)\n\nStraight-through flow rate - balance going Current to 90+ DPD in one quarter. Sudden distress, not normal early delinquency. At 0.26%, approaching the 0.35% pre-GFC level. Sharp climb in the last six months - the early-crack signal stock numbers miss.\n\nSource: NY Fed Consumer Credit Panel / Equifax via Tigzig Tremor platform\n\nNY FED CONSUMER - STUDENT LOANS\n\nAt 10.3%. The 2020-2023 trough was artificial - CARES Act paused federal student loan reporting. Now back on the pre-pandemic trajectory.\n\nSource: NY Fed Consumer Credit Panel / Equifax via Tigzig Tremor platform\n\nU-6 UNEMPLOYMENT\n\nU-6 captures unemployed plus underemployed plus marginally attached. The real labor picture. At 8.2% - back to the pre-GFC 2007 starting level. Jobs side is already where it was before the last credit cycle broke.\n\nSource: BLS via Tigzig Tremor platform\n\nNY FED CONSUMER - HELOC\n\nTapping more, falling behind more.\n\nUtilization at 42.6%, climbing every quarter since Q1-2023. HELOC at 7-8% is the cheapest credit left - with cards at 13% delinquency and 20%+ rates, the house is becoming the emergency fund.\n\nDelinquency at 0.95% - 2.6x the Q2-2024 low. Still below the 1.2% pre-GFC level, but the trajectory is clear.\n\nSource: NY Fed Consumer Credit Panel / Equifax via Tigzig Tremor platform\n\nHE Revolving (HELOC) Utilization\n\nHELOC Balances 90+ DPD (Stock)\n\nHOME PRICES & EQUITY\n\nMedian price flat after the 2022 spike - down from $442.6K to $403.2K. Equity-rich share declining (46.2% to 43.3% YoY), seriously underwater up to 3.2% in 45 states. And the collateral itself is now weakening.\n\nSource: FRED (Median Sales Price) via tremor.tigzig.com and ATTOM Q1-2026 Home Equity & Underwater Report\n\nEquity-Rich Mortgages (% of mortgaged homes)\n\n46.2%\n\nQ1'25\n\n44.6%\n\nQ4'25\n\n43.3%\n\nQ1'26\n\nSeriously Underwater (% of mortgaged homes)\n\n2.8%\n\nQ1'25\n\n3.0%\n\nQ4'25\n\n3.2%\n\nQ1'26\n\nMedian Sales Price of Houses Sold (USD)\n\nWARNINGS\n\nJamie Dimon, CEO JPMorgan Chase\n\n“I do believe that when we have a credit cycle, which will happen one day, losses on all leveraged lending in general will be higher than expected, relative to the environment. This is because credit standards have been modestly weakening pretty much across the board.”\n\n2026 Annual Letter to Shareholders, Apr 2026 · Norges Bank Investment Management Conference, Apr 28, 2026\n\nEquifax Insights\n\n“Credit risk dynamics are shifting as improvements in personal loans and private label delinquency are being countered by a rise in severe mortgage delinquency, alongside a worsening in subprime vintages for both Bankcard and unsecured personal loans through 2024.”\n\nEquifax Insights, April 2026 Consumer Pulse: The Latest Consumer Credit Trends, Dave Sojka, Apr 20, 2026\n\nAuto is past GFC peak. Cards at GFC peak. Mortgage cracking at the flow. The setup is already worse than pre-GFC 2007.\n\nAnd the worst impact of AI labor displacement, oil and inflation shocks, private credit stress and asset overvaluations is yet to play out.\n\nData & Tools\n\nPrimary Data Sources\n\nNY Fed Consumer Credit Panel / Equifax\n\nFDIC Statistics on Depository Institutions\n\nNCUA Quarterly Aggregate FPR\n\nFFIEC Call Reports (NDFI)\n\nBLS - Bureau of Labor Statistics\n\nTREMOR Platform - tremor.tigzig.com\n\n4 modules. ~150 indicators. ~340 quarters of pooled history.\n\n3-layer validation. 1,300+ automated cross-checks.\n\nTime-series, cross-section, compare views.\n\nPer-chart customization. PNG / TSV / CSV exports.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Analysis decks, Quant apps hub, APIs and MCP\n\nMore posts: The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top., Auto Losses Past the GFC Peak. Cards Almost There. Credit Unions Already at GFC Loss Rates. NY Fed + NCUA Data Now on Tigzig., The Crisis in Auto Loans. 2007 was Sub-Prime Mortgage. Is 2026 Auto?, US Credit Unions on TREMOR. Card Charge-Offs at 5.11% - Already Past the 2009 GFC Peak., US Fed Financial Stability Report May 2026. Same Stress Points - Now in Fed Own Survey."
},
{
"slug": "us-consumer-squeeze-real-retail-jun2026",
"title": "May 2026's Record US Retail Sales Number Is Hiding the Pain. Real Spending Hasn't Grown in Five Years. The Consumer Is Squeezed.",
"tags": [
"portfolio-quants"
],
"content": "May 2026's Record US Retail Sales Number Is Hiding the Pain. Real Spending Hasn't Grown in Five Years. The Consumer Is Squeezed.\n\nPublished: 2026-06-18\n\nThe May 2026 record US retail sales headline ($763.7bn, +6.9% YoY) is hiding the pain of a consumer being squeezed - and the squeeze was here long before the Iran war. Strip out inflation and real goods spending has not grown in five years: of the $763.7bn rung up in May 2026, only $605.9bn is real 2021 goods - the other $157.8bn is pure inflation. Households are holding the line by stretching: personal saving rate back to pre-2008 lows, NY Fed HELOC balances up 40% since 2021. The strain is starting to show: University of Michigan consumer sentiment collapsed to 44.8 (near the lowest on record), and 90+ day delinquencies on all consumer debt are back to 3.4% - past the 2007 mark. Real GDP still grows ~2% on services, AI investment and government, so this is a squeeze on the household goods budget, not an economy-wide downturn. The proof is the balance sheet: thinning savings, more borrowing, delinquencies past 2007. Companion piece: jobs market softening since 2024 (unemployment 4.3%, underemployment 8.1%, long-term joblessness near a 3-year high). Cheaper oil is real relief on the price side, but it does not refill a drained savings account or put the long-term unemployed back to work. 6-slide deck with every series live on tigzig.com -> Tremor (retail RSAFS + RRSFS, saving rate BEA, HELOC + delinquency NY Fed, sentiment U-Mich), pullable by API, Excel, or AI agent. Real retail definition: nominal Advance Retail & Food Services Sales deflated by headline CPI (the honest test of volume vs price).\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/us-consumer-squeeze-real-retail-jun2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/us-consumer-squeeze-real-retail-jun2026 with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/USCONSUMERSQUEEZE_JUN2026.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nSlide 1 - Now look beneath it.\n\nThe headline US retail sales hit a record $763.7 billion in May 2026, up 6.9% on the year. The consumer, we are told, is strong.\n\nNow look beneath it. Strip out inflation, and the story turns on its head. What the record number is hiding, in five charts.\n\nSlide 2 - Spending more, buying the same\n\nAdjusted for inflation, real goods spending has not grown in five years. The record is price, not volume.\n\nFive years on, the cart has not grown. Of the $763.7bn rung up in May 2026, only $605.9bn is real 2021 goods. The other $157.8bn is pure inflation.\n\nChart: nominal vs real retail sales since 2021.\n\nSlide 3 - Holding the line by stretching\n\nEven flat real spending has to be funded. Households are doing it by saving less and borrowing against the house.\n\nThe saving rate is back to its pre-2008 lows, and home-equity borrowing has turned up 40% since 2021. The buffer is thin and the house is back in play.\n\nCharts: personal saving rate (%) and home-equity (HELOC) balances ($ billion).\n\nSlide 4 - The strain is starting to show\n\nA consumer leaning on a thinning buffer eventually shows it, in how they feel and in what they cannot pay.\n\nSentiment has collapsed to 44.8, near the lowest on record. And 90+ day delinquencies are back to 3.4% of all consumer debt, past the 2007 mark.\n\nCharts: consumer sentiment (University of Michigan) and 90+ day delinquency on all consumer debt (%).\n\nSlide 5 - The consumer is squeezed, not broken.\n\nThe \"record retail sales\" headline is nominal. Real goods spending has not grown in five years, and per person it has fallen. And yet real GDP still grows about 2%, on services, AI investment and government.\n\nThis is a squeeze on the household goods budget, not an economy-wide downturn. The proof is the balance sheet: a thinning savings buffer, home-equity borrowing turning up, delinquencies back past 2007.\n\nMore dollars at the till, the same goods in the cart, paid for with less saving and more borrowing. A stretched consumer, holding on.\n\nSlide 6 - Sources & method\n\nEvery series in this note is live on tigzig.com -> Tremor: retail sales (nominal & real), the personal saving rate, NY Fed home-equity balances and 90+ day delinquency, and consumer sentiment.\n\nOne place for all of it. Tremor carries 330+ macro and credit indicators in a single dashboard, so you are not hunting across a dozen sites. Pull any series by API, in Excel, or by pointing an AI agent at it.\n\nAgent-first portal. Point Claude Cowork, Claude Code, or any AI agent at tigzig.com: it finds the series, pulls it, charts it and analyzes it for you.\n\n\"Real\" retail definition\n\nReal retail is nominal Advance Retail & Food Services Sales (Census, RSAFS) deflated by headline CPI (CPIAUCSL), published by FRED as RRSFS. Deflating by CPI, not PPI, is the honest test of volume versus price.\n\nSources\n\nRetail & real retail (Census / FRED) · saving rate (BEA) · HELOC & delinquency (NY Fed Household Debt & Credit) · consumer sentiment (University of Michigan). Pulled June 2026, routed via Tremor.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Market tools hub, Analysis archive, Open data and MCP\n\nMore posts: BIS, the Central Bank of Central Banks, Just Red-Flagged the AI Boom. Railway Mania, Roaring 20s, Dotcom - Same Chart., Euro-Area NPLs Look Calm. Split by Bank Size, the Small-Bank Corporate Book Is at 5.16% and Climbing., The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top."
},
{
"slug": "us-credit-data-q2-2026-tremor-sep2026",
"title": "All Five US Credit Datasets on TREMOR Now Run to Q2 2026",
"tags": [
"portfolio-quants",
"database-ai"
],
"content": "All Five US Credit Datasets on TREMOR Now Run to Q2 2026\n\nAll five US credit datasets on TREMOR now run to Q2 2026, with up to 106 quarters of history - covers US banks, credit unions, household debt, insurers' investments and bank loans to non-banks. You can run your own cuts on the interactive tool, or take the data out, as CSV or through the API and MCP for your AI agent.\n\nTool\nTime series for every segment, in tree format.\nOutstanding balance, % share of total loans, then the delinquency and charge-off rates.\nCross-section: for one quarter or compare up to four quarters side by side.\nDelinquency is on each source's own basis. Banks show noncurrent, 90+ and 30+ days, households show 90+ including severely derogatory plus the flow rates, and credit unions show 60+ and 30-59 days.\n\nData\n\nEach dataset has a methodology section on how it is built, and a validations section that reconciles it against the published source and calls out data anomalies.\n\nDownloads\n\nEvery dataset as CSV from the tool. Banks, credit unions and households are also in the full TREMOR database download, and on the API and MCP.\n\nLinks\n\nTool: tigzig.com/tremor - US-NDFI / US-Credit / US-Insurance\n\nAnalyses: tigzig.com/analysis\n\nFor your AI agent: api.tigzig.com\n\nHow the data is built\n\nCovered in detail in the Validations & Methodologies section. Some quick points below.\n\nBanks. The FDIC Call Report data is pulled bank by bank from the FDIC's API and added up to industry totals. The rates are computed from the filed amounts. 30+ is 30-89 days plus 90+ plus nonaccrual, over balance. Noncurrent, FDIC's own headline measure, is 90+ plus nonaccrual. Charge-offs are filed year to date, so the quarter is this quarter's figure minus last quarter's, then annualised. The totals are reconciled against FDIC's Quarterly Banking Profile.\n\nHouseholds. These come from the NY Fed's quarterly Excel, with the rates as the NY Fed publishes them. The 90+ stock includes severely derogatory, meaning balances already charged off that still sit on the credit file. Cross-checked against the NY Fed's own narrative report with zero mismatches.\n\nCredit unions. These are read from NCUA's quarterly Excel reports, with the same year-to-date conversion on charge-offs. Over the years NCUA changed account codes, moved loans between categories, and in a few places filed a year-to-date figure lower than the quarter before. I publish what NCUA filed and explain each of these on the validation page.\n\nMy Q2 2026 notes on this data\n\nFDIC banks. Charge-offs are down to 0.56 percent from a 0.70 percent peak, and lending to other financial firms is now 18.6 percent of all bank loans. FDIC Q2 2026 bank credit\n\nNY Fed households. The headline 90+ flow came down to 2.57 percent, but mortgage and auto went the other way. NY Fed household debt, Q2 2026\n\nCredit unions. Charge-offs are 0.76 percent, above the pre-crisis level, and 60+ day delinquency is above pre-COVID in every segment. Credit unions, Q2 2026\n\nBank loans to non-banks. US banks had 2.8 trillion dollars committed at the end of Q2, and 1.7 trillion of it was already drawn. Bank loans to non-banks, Q2 2026\n\nAll five US credit datasets on TREMOR now run to Q2 2026: banks, credit unions, households, insurers and bank loans to non-banks"
},
{
"slug": "us-credit-unions-ncua-tremor-card-nco-past-gfc",
"title": "US Credit Unions on TREMOR. Card Charge-Offs at 5.11% - Already Past the 2009 GFC Peak.",
"tags": [
"portfolio-quants"
],
"content": "US Credit Unions on TREMOR. Card Charge-Offs at 5.11% - Already Past the 2009 GFC Peak.\n\nPublished: 2026-04-25\n\nUS Credit Unions analytics on TREMOR. 23 years, 86 quarters, $1.72T loans across 4,287 federally-insured credit unions from NCUA quarterly FPR. Card NCO at 5.11% past 4.68% GFC peak. Consumer NCO at 1.82% vs 0.99% pre-crisis. 18 segments, 30-59 DPD, 60+ DPD, NCO. CSV download with 74-field reference.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/us-credit-unions-ncua-tremor-card-nco-past-gfc\nMarkdown of the HTML page: send GET https://tigzig.com/post/us-credit-unions-ncua-tremor-card-nco-past-gfc with header Accept: text/markdown\n\nTags\nportfolio-quants\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, TREMOR Macro Indicators, Quants Agent\n\nExplore: Quant Apps hub, Analysis hub, APIs and MCP\n\nMore posts: Red Flag: US Credit Unions. A $1.73T System at GFC-Era Losses. Card Charge-Offs 10 Straight Quarters Past the Crisis Peak., Auto Losses Past the GFC Peak. Cards Almost There. Credit Unions Already at GFC Loss Rates. NY Fed + NCUA Data Now on Tigzig., Q1 2026 US Lending Data Live on Tigzig. NCUA + FDIC + NY Fed in One API. Two of Three Not on FRED. AI-Agent Ready., Releasing US-CREDIT. For Analysts Tracking US Banking. Interactive Tool Covering FDIC SDI and NY Fed CCP."
},
{
"slug": "us-inflation-shock-no-modern-playbook-jun2026",
"title": "There Is No Modern Playbook for This. US Core Inflation at 3.3%, Re-Accelerating. Producer Prices at 6.4%. Consumer Delinquencies Past 2007.",
"tags": [
"portfolio-quants"
],
"content": "There Is No Modern Playbook for This. US Core Inflation at 3.3%, Re-Accelerating. Producer Prices at 6.4%. Consumer Delinquencies Past 2007.\n\nPublished: 2026-06-12\n\nUS core PCE - the gauge the Fed targets at 2% - is at 3.3% and re-accelerating. Outside Covid, last at this level in April 1992. 25-year pre-Covid average: 1.7%. It has now spent 62 straight months above target. And the pipeline says more is coming: PPI final demand at 6.4% - hottest in the modern series outside Covid - with core PPI at 4.9%; PPI is running 2.2 points ahead of CPI, putting the bigger share of the bill on producers (margin compression) but with consumer pass-through inevitable. The consumer is already stretched: gasoline at $4.65 (only June/July 2022 ever higher), and 3.4% of consumer debt is in the 90+ DPD bucket - past the 3.1% mark of late 2007 and more than double the end-2022 low, rising in 12 of the last 13 quarters. Against all of this the S&P set a record at 7,610 on June 2 with Shiller CAPE 2nd-highest in 145 years and Michigan sentiment lowest in 73 years - a handful of AI names doing most of the lifting. The last two times the market was priced like this while the economy cracked underneath - 2000 and 2007 - the index halved. 6-slide deck with full series codes, method, and the dot-com (-49%) / GFC (-57%) drawdown references.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/us-inflation-shock-no-modern-playbook-jun2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/us-inflation-shock-no-modern-playbook-jun2026 with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/USINFLATIONSHOCK_JUN2026.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nSlide 1 - There is no modern playbook for this\n\nThis is US consumer inflation - core PCE, the measure the Fed targets at 2%. For 25 years it averaged 1.7%. Covid broke that anchor, the gauge never made it back, and it is re-accelerating: 3.3% and rising, before the new oil shock has fully passed through.\n\nOutside Covid, the last time this gauge stood here was 1992 - now 62 straight months above target. Every playbook younger than that assumes the anchor holds.\n\nSlide 2 - Upstream, it is even hotter\n\nProducer prices - what businesses pay before anything reaches a shelf - are rising 6.4% a year. Outside the Covid spike, that is the hottest reading in the modern series. Core PPI, with food and energy stripped out, says the same thing at 4.9%.\n\nWholesale inflation only partly flows downstream - but at 6.4%, even partial pass-through keeps the consumer number rising. The pressure is upstream of the shelf.\n\nSlide 3 - Somebody has to pay\n\nThe bill is always shared - some passes to the shelf, some stays on the producer's books. Right now producer prices are running 2.2 points ahead of consumer prices, so the bigger share sits with the producer: margin compression.\n\nEither margins give, or prices do. Both roads lead through the consumer.\n\nSlide 4 - The consumer is already stretched\n\nWhatever does get passed on hits a household with little slack left. Gasoline at $4.65 - only June and July 2022 were ever higher. And 3.4% of consumer debt is already in that 90+ bucket, past the 3.1% mark of late 2007 and more than double the end-2022 low.\n\nA squeezed paycheck resolves two ways: spend less, which feeds back into growth, or miss payments. The lower line says the second is already happening - rising in 12 of the last 13 quarters.\n\nSlide 5 - Priced as if none of this is happening\n\nAgainst all of the above, the S&P set a record on June 2. Valuations sit near their highest in 145 years - only the dot-com peak ran higher - while consumer sentiment is at its lowest in 73 years of data, and a handful of AI names do most of the lifting.\n\nThe last two times the market was priced like this while the economy cracked underneath - 2000 and 2007 - the index halved. Both are on the chart.\n\nSlide 6 - Previous analysis, sources & method\n\nPrevious analysis - this series\nMarkets at a record while the economy hurts - peak valuations, record-low consumer sentiment. 8 Jun 2026.\nThree red flags hit together - 30Y at 5.14%, PPI heating, Brent above $100. 20 May 2026.\nAre we headed for stagflation-lite? - the Fed caught between inflation and weak jobs. 28 Mar 2026.\n\nThe data tools\nPPI, core PCE, gasoline, the S&P and the NY Fed consumer-credit series are live on tigzig.com -> Tremor - a macro early-warning dashboard, 100+ indicators.\nAgent-first portal with MCP servers and free APIs: agents.tigzig.com - point your AI at it.\n\nSources & method\nSeries: core PCE PCEPILFE · PPI Final Demand PPIFIS · core PPI PPIFES · CPI CPIAUCSL · gasoline APU000074714 · NY Fed 90+ DPD stock · S&P 500 daily. YoY from seasonally adjusted indexes; pulled 11 June 2026.\nCore PCE was last at or above today's 3.3% (outside Covid) in April 1992; above the 2% target every month since March 2021; 1995-2020 average 1.7%.\nThe PPI Final Demand series begins in 2010 and does not cover the 1970s-80s; the Covid window excluded is 2021-2023. Gasoline is in nominal dollars. PPI and CPI baskets differ; pass-through is partial and lagged.\nValuation (Shiller CAPE, 2nd-highest in 145 years) and sentiment (Michigan, lowest in 73 years) figures are from the 8 June note above. S&P drawdowns: dot-com -49% (Oct 2002), GFC -57% (Mar 2009), record 7,610 on 2 Jun 2026.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: Analysis deck series, Market tools hub, APIs and MCP\n\nMore posts: Index at a Record. Peak Valuation. Record-Low Sentiment. Last Two Times This Setup Held - 2000 and 2007 - the Market Halved., Three Red Flags Hit Together. 30Y at 5.14%, PPI at 5.99%, Brent Above $100. Setup Worse Than 2000 or 2007., Are We Headed for Stagflation-Lite, S&P at 7,400. Irrational Exuberance. The Last Two Times This Setup Held, the Index Halved., The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top."
},
{
"slug": "us-insurance-assets-tracker",
"title": "US Insurance Assets Tracker - 25 Years of Data, 73 Series, Life & P&C",
"tags": [
"portfolio-quants"
],
"content": "US Insurance Assets Tracker - 25 Years of Data, 73 Series, Life & P&C\n\nNew tool release - where US insurers invest their money.\n\n73 line items. 104 quarters. 25 years of data. Life and P&C. General Account and Separate Account.\n\nRun your own cuts and charts - by series, period, or share of balance sheet. Download CSV.\n\nLive now: tremor.tigzig.com -> US-Insurance\n\nDid you know?\n\nLife insurers' \"Unidentified Misc Assets\" grew 76% over the last 5 years to $1.2T. Now over 10% of all US Life Insurance assets.\n\nThis is the Fed's catch-all bucket - private credit, PE, hedge funds, BDCs, infrastructure debt, real estate equity. The same sector now under stress from rising defaults and recent frauds.\n\nSources & Methodology\n\nSource is FRED and Fed Z.1 Financial Accounts. Validated against the Fed's L.116 table and NAIC/ACLI Fact Book. Methodology and validation in the app.\n\nUS Insurance Assets Tracker\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Private Credit and NDFI Tracker, Quant tools hub, Analysis hub\n\nMore posts: Red Flag in US Life Insurance. $2.4T (7.5% of GDP) in Private Credit and Offshore Reinsurance., Private Credit Default Rate Hits a Record 6.0% (2026): A Market for Lemons? Banks and Insurers Deep In It., TREMOR. Six Free Tools for Banking and Insurance Analysts. Multi-Layer Validated. Built by a Practitioner., The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top."
},
{
"slug": "us-jobs-calm-before-storm-sep2026",
"title": "On the Headline Numbers the US Job Market Looks Fine. The Story Changes When You Dig Deeper.",
"tags": [
"portfolio-quants"
],
"content": "On the Headline Numbers the US Job Market Looks Fine. The Story Changes When You Dig Deeper.\n\nOn the headline numbers the US job market looks fine. The story changes when you dig deeper. Taken together with the pressure on the economy, markets and geopolitics, unless something changes drastically, I believe we are heading towards a tipping point. And not only for the US economy.\n\nThe August jobs report came in well above expectations... and twelve days later the Fed raised rates and described the economy as growing at a solid pace.\n\nA deeper analysis gives a more mixed picture. Job growth has slowed from 3.2 per cent a year in January 2023 to 0.4 per cent now, and payroll jobs for August 2024 and August 2025 were each revised down by about a million. Hiring is near its lowest on record, and people who lose a job are staying out of work longer, with the median time out of work up from 9.6 weeks to 11.4.\n\nHourly pay has grown more slowly than prices for five months in a row, and the saving rate is down to 3.0 per cent as spending keeps running ahead of income. Some measures point the other way. Layoffs are low and weekly pay is still just ahead of prices, and the deck shows those as well.\n\nLabour is one part of a bigger story. Links to earlier analysis are in the deck and below.\n\nA few things about the labour data that are easy to trip on\nThe numbers come from several surveys, and each counts something different. Payrolls count jobs and the household survey counts people, so someone with two jobs counts twice in one and once in the other. A payroll change has to reach about 122,000 before it stands out from survey noise, and a household change about 650,000.\nThe household survey resets its population estimates every January. This January the reset took 0.4 of a point off the participation rate in one month, so any comparison across January needs care.\nPayrolls are checked once a year against tax records. The early estimate of that revision comes out around late August and the final one in February, and the August counts for 2024 and 2025 were each revised down by about a million.\nLayoffs show up in two ways. JOLTS layoffs count people let go in a month, and that number is flat. Permanent job losers count people laid off and still out of work, and that number has gone up because people are taking longer to find work.\nOctober 2025 is missing from the household survey, because the survey was not carried out during the government shutdown.\n\nSome of the earlier analysis on the economy, markets and valuation risks\nBond markets are sounding an alarm. The US thirty year went above 5.2 per cent this year for the first time since 2007, and long rates kept rising even while the Fed was cutting. Bond markets are sounding an alarm\nUS banks have 2.8 trillion dollars committed to non-bank lenders, and 1.7 trillion of it is already drawn, while Fitch has private credit defaults at a record. US banks and non-bank lending, Q2 2026\nCredit unions in Q2 2026. Losses are above the pre-crisis level, and delinquency is above pre-COVID in every segment. Credit unions, Q2 2026\nA correction in AI valuations looks more likely. A central bank, an asset manager and a bond investor now say so, and it all rests on free cash flow arriving from 2028. A correction in AI valuations\nThe FSB chair is not hedging. Andrew Bailey told the G20 that markets remain vulnerable to a disorderly correction that could spread across borders. The FSB chair's warning\n\nAll of them are at tigzig.com/analysis\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals are in the PDF and slide images on the HTML page.\n\nSlide 1\n\nTIGZIG\n US · MACRO ANALYSIS\n 24 SEPTEMBER 2026\n\n \n US jobs: the calm before the storm?\n\n The unemployment rate, at 4.1 per cent, and some other\n measures have eased this year. The Fed calls the economy solid, and several\n economists called August's jobs report strong.\n But the pressure built over three years is still there: people are out of work for\n longer, hiring is near its lowest on record, and payroll jobs for August 2024 and\n 2025 were each revised down by about a million. And the pressure from outside the job\n market has not let up.\n\n THE HEADLINE NUMBER\nUnemployment rate\nper cent\n 3.54.1\nJan 2023Aug 2026\n\nPayroll jobs, YoY\nper cent change on a year\n 3.20.4\nJan 2023Aug 2026\n\nOut of work 6 months+\nper cent of the labour force\n 0.661.14\nJan 2023Aug 2026\n\nMedian time out of work\nweeks\n 9.611.4\nJan 2023Aug 2026\n\nPersonal saving rate\nper cent\n 4.93.0\nJan 2023Jul 2026\n\nHiring rate\nper cent of employment\n 4.13.2\nJan 2023Jul 2026\n\n SourcesFRED, BLS and BEA. Full sources on page two and each page.\n\n Amar Harolikar, ACA · Decision Sciences & Applied AI\n\n 1 of 16\n\n tigzig.com\n\nSlide 2\n\nTIGZIG\n US · MACRO ANALYSIS\n 24 SEPTEMBER 2026\n\n \n\nWhere these numbers come from\n\n Six official sources carry most of this deck, each counting a\n different thing on a different clock.\n\n \n\nWhy the two job numbers disagree\n\n Payrolls count jobs. The household survey counts people. Someone holding two jobs adds two to the first and one to the second.\nThe household survey is the wider one. It takes in the self-employed, farm work, paid help in homes and people on unpaid leave. Payrolls leave all four out.\nNeither one is exact. A payroll change has to reach 122,000 before the survey can tell it from noise. A household change has to reach 650,000.\n\n Every series here can be pulled from its original source. Many are\n also in one place on TIGZIG\n TREMOR, to chart, overlay, download or pull through an API or MCP.\n\n| Source | Who is counted | How late | Is it revised |\n| --- | --- | --- | --- |\n| CPSThe household survey | About 60,000 households. Counts people, so a second job shows once. | First Friday after the month | Once a year. The seasonally adjusted series, used here, are recut each January over five years. |\n| CESThe payroll survey | 119,000 businesses, 622,000 worksites, 26 per cent of payroll jobs. Counts jobs. | Same day as the household survey | Twice in two months, then checked each year against a near census of employers. |\n| JOLTSOpenings and turnover | About 21,000 establishments. | Five to six weeks after | Once, the next month. Then pinned to revised payrolls each January. |\n| CPIConsumer prices | Prices taken in 75 urban areas. Urban households, no farm or rural. | About two weeks after | The raw index, never. The seasonally adjusted series, used here, is recut each year over five years. |\n| BEAThe national accounts | Nobody. An accounting build from Census, BLS, IRS and Treasury data. | About four weeks after | Three cycles. Each month, five years every September, a rebuild every five years. |\n| ALFREDEvery figure as first published | Not a survey. The archive of what each series said on the day it came out. | Updated with each release | Never. That is the point: it keeps the versions others overwrite. |\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 2 of 16\n\n tigzig.com\n\nSlide 3\n\nTIGZIG\n US · MACRO ANALYSIS\n 24 SEPTEMBER 2026\n\n \n\nThe Fed and several economists read the same numbers as solid\n\n The Fed raised rates, and economists called August strong.\n\n The Fed, 16 September\n\n What the Fed says\nWhat the data shows\n\n \"Economic activity is expanding at a solid pace.\"\nReal GDP is up 2.1 per cent on a year earlier.\n\n\"Job gains have kept pace with the workforce, and the unemployment rate has changed little.\"\nThe rate has held between 4.0 and 4.5 since 2025. But the workforce is 688,000 smaller than in January, so keeping pace takes fewer jobs.\n\n Economists and strategists on August jobs, 4 September\n\n Payrolls rose 162,000 against about 53,000\n expected, the most since March.\n\n \"This print was unambiguously strong.\"\nJeff Schulze, Franklin Templeton\n\n\"...evidence of a stable labor market...\"\nJennifer Timmerman, Wells Fargo\n\n\"Very strong\" on payroll.\nJan Hatzius, Goldman Sachs\n\nLabor market clearly not weakening\nRoger Ferguson, former Fed vice chair\n\n Not all agreed. \"The broader trend of labor market rebalancing is\n still intact,\" said Tim Urbanowicz of Innovator ETFs. The next pages look at that\n side.\n\n SourcesFederal Reserve, FOMC statement, 16 Sep 2026. CNBC, 4 Sep 2026: Schulze, Timmerman, Urbanowicz and the Dow Jones payroll forecast from CNBC's round-up of the report; Hatzius and Ferguson as CNBC headlined their interviews (Hatzius, Ferguson). Data from FRED; labour force January to Aug 2026 on one population base.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 3 of 16\n\n tigzig.com\n\nSlide 4\n\nTIGZIG\n US · MACRO ANALYSIS\n 24 SEPTEMBER 2026\n\n \n\nThe headline numbers look calm, but jobs are being added far more slowly\n\n Job growth has slowed sharply over three years. This year has been\n better than last, ▶ in months not yet revised.\n\n Payroll jobs millions Jan 2023 Aug 2026 154.8 159.1 ▶\nJobs are at a record 159.1 million.\nYoY growth: 3.2 per cent in Jan 2023, 0.07 in Dec 2025, 0.4 now.\nAdded Jan to Aug: 1.9 million in 2023, 900,000 in 2024, 156,000 in 2025, 643,000 in 2026.\nFive months of 2025 lost jobs, four within the margin of error. 2026 has had one, in February.\nUnemployment is 4.1 per cent: 3.5 in Jan 2023, 4.3 in January.\n\n Unemployment rate per cent Jan 2023 Aug 2026 3.5 4.1 ▶ Jobs added each month thousands, change on month Jan 2023 Aug 2026 Payroll jobs, YoY per cent change on a year Jan 2023 Aug 2026 3.2 0.4 ▶\n\n SourcesFRED. Unemployment rate from the Current Population Survey, a survey of households; payroll jobs from the Current Employment Statistics survey, a survey of employers; both from the US Bureau of Labor Statistics, seasonally adjusted. The bars are the change on the previous month and the latest two are preliminary. YoY is the per cent change on the same month a year earlier.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 4 of 16\n\n tigzig.com\n\nSlide 5\n\nTIGZIG\n US · MACRO ANALYSIS\n 24 SEPTEMBER 2026\n\n \n\nPayroll jobs were revised down by a million, two years running\n\n 159,005 First Nov 2024 159,061 Second Dec 2024 159,068 Third Jan 2025 158,358 True-up Feb 2025 157,945 True-up Feb 2026 Payroll jobs for October 2024, counted\n five times, in thousands.\n\nThe last\n count is 1,060,000 fewer than the first. Both big cuts came at the yearly\n true-ups, against tax records.\n\n \n Thousands\n\n First\n\n Today\nChange\n\n August 2021\n147,190\n147,314\n+124 (+0.08%)\n\nAugust 2022\n152,744\n153,362\n+618 (+0.40%)\n\nAugust 2023\n156,419\n156,261\n-158 (-0.10%)\n\nAugust 2024\n158,779\n157,757\n-1,022 (-0.64%)\n\nAugust 2025\n159,540\n158,472\n-1,068 (-0.67%)\n\nAugust 2026\n159,075\n159,075\nnot yet\n\n It does not always go this way. Fourteen of the twenty-six\n Augusts since 2000 now stand higher than first reported. The early estimate of the\n next true-up, published on 28 August, is a cut of 79,000, far smaller\n than the last two.\n\n SourcesALFRED, the vintage archive behind FRED, which holds every figure as first published. Payroll jobs, seasonally adjusted, from the Current Employment Statistics survey (CES), US Bureau of Labor Statistics. The staircase is October 2024 counted five times; the table is August of each year. The yearly true-up reconciles the survey against tax records, and August 2026 has not had one.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 5 of 16\n\n tigzig.com\n\nSlide 6\n\nTIGZIG\n US · MACRO ANALYSIS\n 24 SEPTEMBER 2026\n\n \n\nUnemployment fell this year, but people are out of work for longer\n\n Fewer people are out of work than at their recent peaks. Those who\n are out are staying out longer than three years ago, and that has barely\n eased. Since January 337,000 fewer are unemployed, but 351,000 fewer are in work too, on\n the household survey. Payrolls show 483,000 more jobs.\n\n How many are out of work\n\n Unemployment rate per cent Jan 2023 Aug 2026 3.5 4.1 ▶ Broad rate, U-6 adds part-timers, discouraged Jan 2023 Aug 2026 6.7 7.7 ▶ Laid off for good per cent of the labour force Jan 2023 Aug 2026 0.78 1.05 ▶\n\n How long they stay out of work\n\n Out of work 15 weeks+ per cent of the labour force Jan 2023 Aug 2026 1.2 1.8 ▶ Out of work 6 months+ per cent of the labour force Jan 2023 Aug 2026 0.66 1.14 ▶ Median time out of work weeks Jan 2023 Aug 2026 9.6 11.4 ▶\n\n SourcesFRED, Current Population Survey, published by the US Bureau of Labor Statistics, seasonally adjusted. The unemployment rate is U-3, the published headline. U-6 adds people working part time who want more hours and people who have stopped looking. Out of work 15 weeks+ is U-1. Laid off for good is permanent job losers still unemployed, a count of people out of work now; JOLTS layoffs on page eight count layoffs in a month. There is no October 2025 reading in any of these series: the survey was not carried out during the government shutdown.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 6 of 16\n\n tigzig.com\n\nSlide 7\n\nTIGZIG\n US · MACRO ANALYSIS\n 24 SEPTEMBER 2026\n\n \n\nJob openings are high, but hiring is near its lowest on record\n\n Hiring has been lower than today only during and just after the\n 2008 financial crisis, for one month in the pandemic, and in February this\n year.\n\n Job openings 4.4 per cent of employment plus openings 2 3 4 5 6 7 2004 2014 Jul 2026 Hires 3.2 per cent of employment peaks at 6.1 in May 2020, above the top 3.0 3.5 4.0 4.5 2004 2014 Jul 2026\n\n Employers are still advertising, but hiring has slowed to\n levels last seen after 2008.\n\n SourcesFRED. The Job Openings and Labor Turnover Survey (JOLTS), published by the US Bureau of Labor Statistics, monthly since December 2000. Openings have been lower than today's 4.4 in 226 of 308 months, hires lower than 3.2 in 21. The two rates do NOT share a denominator: openings are a share of employment plus openings, hires a share of employment alone, so they can be read side by side and cannot be subtracted from one another. Ends in Jul 2026.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 7 of 16\n\n tigzig.com\n\nSlide 8\n\nTIGZIG\n US · MACRO ANALYSIS\n 24 SEPTEMBER 2026\n\n \n\nPeople are holding on to the jobs they have\n\n Layoffs have been lower only in 2021 and 2022, when employers\n were short of workers, and in one month of 2024. People quit about as rarely as in\nLayoffs 1.0 per cent of employment peaks at 8.6 in Mar 2020, above the top 1.0 1.5 2.0 2.5 2004 2014 Jul 2026 Quits 1.9 per cent of employment 1.5 2.0 2.5 3.0 2004 2014 Jul 2026\n\n Slow hiring normally comes with more people losing their jobs.\n Not this time. Hiring and layoffs this low together did not happen once from 2000\n to 2023, and it has happened 13 times since June 2024.\n\n If layoffs are this low, why is \"laid off for good\" on page\n six higher than three years ago? Layoffs count people let go each month, and\n that is flat. Laid off for good counts those still out of work. With hiring slow they\n stay out longer: the average for everyone unemployed went from 20.4 weeks in January 2023 to 26.3.\n\n SourcesFRED. The Job Openings and Labor Turnover Survey (JOLTS), published by the US Bureau of Labor Statistics, monthly since December 2000. Both rates are a share of employment, and the layoffs series is layoffs and discharges. HOW THE MONTHS ARE COUNTED: hiring at or below 3.3 and layoffs at or below 1.1, a tenth above today's readings so one-decimal rounding cannot decide it, with 2020 excluded because both series leave their range that year. On today's exact readings the count is 2. Ends in Jul 2026.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 8 of 16\n\n tigzig.com\n\nSlide 9\n\nTIGZIG\n US · MACRO ANALYSIS\n 24 SEPTEMBER 2026\n\n \n\nBank of America: the stock boom is fuelling a retirement party\n\n The share of over-55s in the workforce fell from 38.6 per cent in\n August 2024 to 37.2. If stocks fall, some may come back.\n\n All adults 61.6 per cent in the workforce, 16 and over 62 64 66 2002 2014 Aug 2026 Over 55 37.2 per cent in the workforce 32 36 40 2002 2014 Aug 2026\n\n \"Labor force participation is collapsing among older workers. We\n think the strength of the equity market is partly to blame.\"\n Bank of America economists, via CNBC\n\n \"An end to the boom-fueled retirement party could pose a risk to\n the labor market and economy.\"\n CNBC, summarising economists\n\n Age plays the larger part. More than 4 million Americans turn 65 each\n year, and people aged 55 to 64 are more likely to be in the workforce than in\nSources and methodUS Bureau of Labor Statistics, Current Population Survey, seasonally adjusted via FRED; charts cross yearly population revisions, and 0.6 of the over-55 fall came in January 2026 when BLS revised its population figures. Age: a shift-share split by band (55 to 64, 65 and over), January to Aug each year, puts most of the fall since 2019 on the rising share of over-65s. CNBC, 21 Sep 2026: stock gains fuel retirements (also the 4 million).\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 9 of 16\n\n tigzig.com\n\nSlide 10\n\nTIGZIG\n US · MACRO ANALYSIS\n 24 SEPTEMBER 2026\n\n \n\nPay growth slowed, prices sped up\n\n Over the year to Aug 2026, pay growth fell from 4.0 per cent to 3.1 while\n inflation went from 2.9 to 3.4. The two lines crossed.\n\n -2 +0 +2 +4 +6 +8 Pay growth Price growth Pay 3.1% Prices 3.4% 2008 2011 2014 2017 2020 2023 2026 \n What moved the gap, over the year\n\n Pay growth slowed\n0.89 points\n\n Prices sped up\n0.41 points\n\n The gap moved\n1.30 points\n\n SourcesFRED. Average hourly earnings for all private employees, from the Current Employment Statistics survey (CES), against CPI for all urban consumers, year on year, both published by the US Bureau of Labor Statistics. The chart draws both from 2008.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 10 of 16\n\n tigzig.com\n\nSlide 11\n\nTIGZIG\n US · MACRO ANALYSIS\n 24 SEPTEMBER 2026\n\n \n\nSo pay fell behind prices\n\n Private-sector hourly pay has grown more slowly than prices for\n five months in a row, the first such stretch since 2023, by 0.3 of a point\n in the latest month.\n\n How far pay growth ran ahead of or behind price growth, percentage\n points · all private employees, monthly, with Brent below on the same\n timeline. Shaded stretches are the five episodes, listed on the next page\n\n -4 +4 +8 0 1 2 3 4 5 50 100 Brent, dollars a barrel 2008 2012 2016 2020 2024\n\n Sources and methodFRED. Average hourly earnings for all private employees, from the Current Employment Statistics survey (CES), against CPI for all urban consumers, both from the US Bureau of Labor Statistics. Oil is Brent, monthly averages, published by the US Energy Information Administration and taken from FRED. This is hourly pay against headline prices, and both choices decide the answer: against core prices pay is ahead in all five months, and against weekly pay there is no 2026 stretch. The 2020 spike reflects who was still employed, not pay rises.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 11 of 16\n\n tigzig.com\n\nSlide 12\n\nTIGZIG\n US · MACRO ANALYSIS\n 24 SEPTEMBER 2026\n\n \n\nFive times in twenty years\n\n What else was going on. Oil rose in the year before all five, twice as\n a rebound from a crash rather than a shock.\n\n 1\nOct 2007\nto Oct 200813 months below, deepest -2.4 points\n\nOil went from 82 dollars to 133. The financial crisis ran through the same months: recession from December 2007, Lehman in September 2008.\n\n2\nMar 2010\nto May 20103 months below, deepest -0.4 points\n\nGasoline fell 43 per cent in 2008 and came back 54 in 2009, a rebound rather than a shock. Unemployment near 10.\n\n3\nFeb 2011\nto Apr 201215 months below, deepest -1.9 points\n\nThe Arab Spring. Libya's civil war cut its oil output by 60 to 90 per cent and oil went from 104 dollars to 123.\n\n4\nApr 2021\nto Apr 202325 months below, deepest -3.6 points\n\nReopening and supply-chain shortages, then Russia invaded Ukraine. Prices peaked at 9 per cent and core prices at 6.6, which is food and energy stripped out.\n\n5\nApr 2026\nto Aug 2026 now5 months below, deepest -0.8 points\n\nA war with Iran shut the Strait of Hormuz. Oil left its old range in March: above 90 dollars on seven trading days in ten since, against a 2024 to 2025 median of 75.\n\n Sources and methodA stretch is three or more months in a row with pay behind prices, and two stretches three months or less apart count as one; the month count is the months below, not the span. Oil is Brent (DCOILBRENTEU), published by the US Energy Information Administration and taken from FRED; the Libya output cut is from EIA, 2011. Dollar figures are monthly averages; seven in ten is 101 of 142 daily closes since 1 March; the latest months are not final. Gasoline, pay and prices are BLS. Prices here are CPI, and that decides the count: against core PCE, the Federal Reserve's own target index, there is one episode in twenty years rather than five.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 12 of 16\n\n tigzig.com\n\nSlide 13\n\nTIGZIG\n US · MACRO ANALYSIS\n 24 SEPTEMBER 2026\n\n \n\nIncomes have stopped growing and people are saving less\n\n Income per head is up 0.2 per cent over the year, and\n spending per head 1.9. Spending has kept going, so people save less.\n\n The same three years, to Jul 2026\n\n Income per person dollars a year, after inflation Jan 2023 Jul 2026 50,353 52,849 ▶ Spending per person dollars a year, after inflation Jan 2023 Jul 2026 46,167 49,288 ▶ Personal saving rate per cent Jan 2023 Jul 2026 4.9 3.0 ▶\n\n On today's figures, which BEA revises heavily, Americans have\n saved this little in only three other stretches since 1959: a single month in 2001, three months in 2022, and the three\n years to April 2008, before the financial crisis.\n\n Over the year, spending per head grew 1.7 points faster\n than income per head. That gap is why saving fell.\n\n SourcesFRED. Saving rate and real disposable income per head, real consumer spending and population all from the Bureau of Economic Analysis. Spending per head is real personal consumption divided by population, built on the same basis as income per head. All three charts are drawn to Jul 2026, the latest month all three have. THE SAVING RATE IS A RESIDUAL between two much larger numbers and BEA revises it heavily: July 2024 was first published at 2.9 and now reads 5.3. The direction holds on any vintage; a single month should not be read too closely.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 13 of 16\n\n tigzig.com\n\nSlide 14\n\nTIGZIG\n US · MACRO ANALYSIS\n 24 SEPTEMBER 2026\n\n \n\nOther measures give a more mixed picture\n\n The first two point the same way as the pages before. The other four\n show a steadier job market: fewer people stuck in part-time work, few being let go,\n and weekly pay still ahead of prices.\n\n Job openings per unemployed person Jan 2023 Jul 2026 1.79 1.05 ▶ Fewer openings for each person looking, and below the 1.16 of 2018 and 2019.\n\n Continuing claims thousands, still on benefit Jan 2023 Aug 2026 1604 1778 ▶ More people still drawing unemployment benefit, though fewer than last year.\n\n Part time, want full time millions Jan 2023 Aug 2026 4.0 4.4 ▶ Below the 4.6 million of 2018 and 2019, and down from 4.8 a year ago.\n\n New claims thousands a week Jan 2023 Aug 2026 204 206 ▶ No lasting rise, and below the 219 thousand of 2018 and 2019.\n\n Pay minus prices weekly pay, points Jan 2023 Aug 2026 -1.53 0.34 ▶ Ahead of prices from June 2023. Behind in April and May, just ahead now.\n\n Hourly pay minus core points Jan 2023 Aug 2026 -1.05 0.64 ▶ Against prices without food and energy, ahead every month since August 2023.\n\n Sources and methodFRED, from the US Bureau of Labor Statistics and the Department of Labor. Job openings per unemployed person is JOLTS openings over the unemployed, to Jul 2026. Claims are weekly filings to state programmes, averaged by month. Part time, want full time is people working part time for economic reasons. Pay minus prices is year-on-year growth in average weekly earnings (CES0500000011) less CPI inflation; the core panel is average hourly earnings (CES0500000003) less core CPI. All charts run from Jan 2023, with no October 2025 household reading (shutdown).\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 14 of 16\n\n tigzig.com\n\nSlide 15\n\nTIGZIG\n US · MACRO ANALYSIS\n 24 SEPTEMBER 2026\n\n \n\nMy read\n\n Labour is under pressure, and that is one part of a bigger story.\n The pressure runs across the economy, the markets and geopolitics, and the pieces are\n listed on the next page. My read is that, unless something changes drastically, we\n are moving fairly rapidly towards a tipping point, and not only for the US economy.\n\n \n Long rates are going up, and the Fed raised rates in\n September. The US thirty year\n is 5.3 per cent, a level last seen in 2007, and it sets mortgages, refinancing and\n government borrowing alike.\n\n Credit is under pressure, and deteriorating in parts. The\n household, bank and credit union data I track each show it differently.\n\n Private credit is the new risk, and not only for banks. Banks\n have 2.8 trillion dollars committed to non-bank financial firms, Fitch puts defaults at a\n record 6.3 per cent, and life insurers hold a great deal of it.\n\n The AI build has turned cash negative. Three of the five\n hyperscalers had negative free cash flow in the June quarter.\n\n Valuations are close to their record. The Shiller CAPE is 40.6,\n higher in only nineteen months since 1881. It was 27 before the 2008 crash.\n\n Inflation has not gone and the war keeps oil up. Core PCE is\n 3.34 per cent against a 2 per cent target, and Brent has been above 90 dollars on\n seven trading days in ten since March.\n\n The people who run the money are saying it too. Dimon says a credit\n recession would be worse than people think, and the FSB chair told the G20 that leverage and AI\n cross-investment could amplify a correction.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 15 of 16\n\n tigzig.com\n\nSlide 16\n\nTIGZIG\n US · MACRO ANALYSIS\n 24 SEPTEMBER 2026\n\n \n\nMy previous analyses\n\n Each pressure behind this story has its own piece.\n\n Macro and rates\nSep 2026\nBond markets are sounding an alarm\n\nSep 2026\nMacro charts back at their 2007 levels\n\nSep 2026\nThe Fed is caught\n\nConsumer and bank credit\nSep 2026\nCredit unions, losses above pre-crisis\n\nAug 2026\nNY Fed household debt, Q2 2026\n\nPrivate credit and non-bank lending\nAug 2026\nUS banks, 2.8 trillion committed to non-banks\n\nMay 2026\nPrivate credit. A market for lemons\n\nMay 2026\nRed flag in US life insurance\n\nValuations and the AI build\nAug 2026\nA correction in AI valuations looks more likely\n\nJul 2026\nThree warnings on the AI valuation boom\n\nMay 2026\nS&P at 7,400. Irrational exuberance\n\nWhat the regulators are saying\nSep 2026\nThe FSB chair is not hedging\n\nJun 2026\nThe BIS red-flagged the AI boom\n\n All of my analysis\n tigzig.com/analysis\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 16 of 16\n\n tigzig.com\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QDesk - Quant Report Desk, MFPRO - Mutual Fund Analytics\n\nExplore: Analysis hub, The deck as a PDF, Private credit hub\n\nMore posts: Bond Markets Are Sounding an Alarm. The US Thirty Year Is at 5.25 Per Cent and the Fed Has Been Cutting Through All of It., US Banks Had $2.8 Trillion Committed to Non-Bank Lenders at the End of Q2 2026, and $1.7 Trillion of It Was Already Drawn., Credit Union Q2 2026: Losses Above Pre-Crisis, and Delinquency Above Pre-COVID in Every Segment, The FSB Chair Is Not Hedging. Andrew Bailey's Blunt Warning Note to the G20., US Unemployment Is 4.3%. The Stuff Underneath Says Otherwise. 2M Jobless Over Six Months, Up 3 Straight Years."
},
{
"slug": "us-jobs-september-2026-update-oct2026",
"title": "US Jobs, September 2026: Fewer New Jobs, Longer Searches, Pay Barely Keeping Up",
"tags": [
"portfolio-quants"
],
"content": "US Jobs, September 2026: Fewer New Jobs, Longer Searches, Pay Barely Keeping Up\n\nThe US added 29,000 jobs in September. The past year adds up to 543,000 jobs, down from 827,000 as first reported. First reports have been revised down four years in a row, the longest run in records back to 1956.\nFewer workers are joining the labour force. More people are retiring and fewer immigrants are arriving. Reuters reports economists put the jobs needed each month at about 50,000. RBC puts it at 20,000. The last three months averaged 51,000.\nHealth care and social assistance have added more jobs than the whole economy over the past two years. Outside them, monthly job gains fell from 126,000 in 2023 to about zero. One month drags that down. Federal payrolls fell about 166,000 in October 2025 as staff who took a 2025 resignation offer came off the payroll. Leave that month out and jobs outside health care grew about 15,000 a month.\nUnemployment is 4.2 per cent against 3.6 on average in 2023. It has eased this year because fewer people are in the labour force.\nPeople stay out of work longer. The typical spell is 11.5 weeks. 27 per cent of the unemployed have been out for six months or more.\nHiring is low. About 4 in 10 people on unemployment benefit now reach the end of it, against 1 in 3 in 2019.\nHourly pay has trailed prices since April, mostly because of energy. The saving rate is down to 4.1 per cent.\n\nThe first page rates 11 measures against 2023 and 2018-19. Eight are weak, one is mixed and two are strong. The rest of the analysis has the charts behind each one, plus a few more measures.\n\nThis updates my analysis of 24 September 2026:\n\"On the headline numbers the US job market looks fine. The story changes when you dig deeper\".\nhttps://www.tigzig.com/post/us-jobs-calm-before-storm-sep2026\n\nOther analysis at: https://www.tigzig.com/analysis\n\nUse of AI, and a few notes on the numbers\n\nParts of this work were prepared with AI and steered by me. I reviewed, edited and approved every page. Every data series is pulled from the sources above. The analysis and conclusions are mine.\nThe 827,000 adds up each month's job change as it was first reported. The February benchmark revision accounts for about 16,000 of the cut. The rest came from the regular monthly revisions. September is a first estimate and will be revised twice more.\nThe revision history compares the same September to August window in every year, using the jobs report published that October. Later benchmark revisions cut 2024 and 2025 to about 100,000 a month. This year's benchmark comes in February 2027, and the BLS early estimate is a small cut of 79,000.\nSeptember may be understated. Reuters reports economists noting that payrolls tend to come in weak when Labor Day falls late, as it did this year. Reuters also links the July and August revisions to the seasonal adjustment model.\nJOLTS, the openings, hiring, quits and layoffs data, is published a month after the jobs report, so those figures end in August.\nBEA revised income and saving on 30 September. The saving figures here are the revised ones.\n\nMy earlier analyses on the economy and markets\nThe labour deck from 24 September. On the headline numbers the job market looked fine, and the numbers underneath told a different story.\nhttps://www.tigzig.com/post/us-jobs-calm-before-storm-sep2026\nThe bond market rout. The US ten year is above 5 per cent for the first time since 2007, and those rates set mortgages, company refinancing and government borrowing.\nhttps://www.tigzig.com/post/bond-market-rout-sep2026\nUS diesel went above $6 a gallon for the first time, and the cost is reaching truck freight, airfares and farm bills.\nhttps://www.tigzig.com/post/diesel-above-six-dollars-sep2026\nCredit union losses in Q2 2026 are above the pre-crisis level, and 60 plus day delinquency is above pre-COVID in every segment.\nhttps://www.tigzig.com/post/credit-union-q2-2026-losses-delinquency-sep2026\nA Correction in AI Valuations Is Looking More Likely\nhttps://www.tigzig.com/post/ai-valuation-correction-aug2026\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals are in the PDF and slide images on the HTML page.\n\nSlide 1\n\nTIGZIG\n US JOBS · SEPTEMBER 2026\n 4 OCTOBER 2026\n\n \n Fewer new jobs, longer searches, pay barely keeping up.\nLayoffs and new claims are low. With more people retiring and fewer immigrants arriving,\n economists say fewer new jobs are needed. Most other measures are worse than in 2023 and before the pandemic.\nJobs added in the 12 months to August: 827K as first reported, revised down\n 34% to 543K. First reports have been revised down four years in a row, the longest run\n in records back to 1956.\n\n Measure\nLatest\n\nStatus\nLast 3 months\n\nJobs added, 3-month average\n51K\nSep\nWeak\n▼Worse\n\nUnemployment rate\n4.2%\nSep\nWeak\n▲Better\n\nTypical time out of work, weeks\n11.5\nSep\nWeak\n▶Steady\n\nUnemployed 6 months+, share\n27.1%\nSep\nWeak\n▼Worse\n\nJob openings per person looking\n1.01\nAug\nWeak\n▲Better\n\nHiring rate\n3.3%\nAug\nWeak\n▶Steady\n\nLayoffs rate\n1.0%\nAug\nStrong\n▲Better\n\nNew jobless claims, a week\n200K\nSep\nStrong\n▲Better\n\nHourly pay minus prices\n-0.2 pts\nAug\nWeak\n▼Worse\n\nWeekly pay minus prices\n+0.4 pts\nAug\nMixed\n▼Worse\n\nSaving rate\n4.1%\nAug\nWeak\n▼Worse\n\n Status compares the latest reading with the 2023 and 2018-19 averages. Weak\n is worse than both, mixed is worse than one, strong is better than both. Last 3 months is our read of each chart:\n the last three months against January to March 2026.\n\n SourcesBLS, Department of Labor and BEA, via FRED. Full sources and method on the last page.\n\n Amar Harolikar, ACA · Decision Sciences & Applied AI\n\n 1 of 11\n\n tigzig.com\n\nSlide 2\n\nTIGZIG\n US JOBS · SEPTEMBER 2026\n 4 OCTOBER 2026\n\n \n The past year was smaller than first reported.\n\n \n Jobs added, 12 months to August\n 827K First reported 543K Today\n\n \n Revised down 34%\n\nAdded up month by month as each was first reported, the 12 months to August came to\n 827K jobs. Today the same months add up to 543K. 8 of the 12 were revised down.\n\n How many jobs a month does the US need now? Fewer, because\n retirements and the immigration crackdown mean fewer new workers. Reuters reports\n economists put this break-even at about 50K a month. RBC puts it at 20K. The last three\n months averaged 51K.\n\n The same 12 months, plus September, in thousands\n\n First reported As published today\n · September is a first estimate\n\n +76 Sep 25 -140 Oct +41 Nov -17 Dec +160 Jan 26 -156 Feb +214 Mar +148 Apr +63 May +31 Jun -10 Jul +133 Aug +29 Sep 26\n\n SourcesBLS payroll survey via ALFRED and FRED; break-even, Reuters and RBC. Full sources and method on the last page.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 2 of 11\n\n tigzig.com\n\nSlide 3\n\nTIGZIG\n US JOBS · SEPTEMBER 2026\n 4 OCTOBER 2026\n\n \n Job growth has been revised down four years in a row. That is the longest\n run in records back to 1956.\n\n How much each year's jobs were revised, thousands a month\n\n Revised up Revised down\n · 12 months from September to August\n\n -50 -25 +25 +50 +75 +100 0 1980 1990 2000 2010 2020 2026 4 years down in a row\n\n \n The cuts usually grow later. Each bar compares first reports with that\n October's jobs report. Every February, BLS checks the past year against tax records.\n That took 2024 and 2025 to cuts of about 100K a month, the second and fourth largest since\nThis year's comes in February 2027, and the early estimate is\n a small cut of 79K.\n\n MethodEach bar is the 12-month change in that year's October report, minus the sum of the months as first reported, divided by 12. ALFRED, every vintage of PAYEMS, which go back to 1956; before 1980 no run was longer than one year. The September 2025 report, normally out in early October, came on 20 November after the shutdown.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 3 of 11\n\n tigzig.com\n\nSlide 4\n\nTIGZIG\n US JOBS · SEPTEMBER 2026\n 4 OCTOBER 2026\n\n \n Job gains fell from 210K a month in 2023 to 41K. Outside health care, from\n 126K to near zero.\n\n Average jobs added a month, in thousands\n\n All jobs +210 2023 +41 Last 12m\n Health care, social +83 2023 +43 Last 12m\n All except health, social +126 2023 -2 Last 12m +15 Excl. Oct 25\n Government +60 2023 -18 Last 12m\n Professional, business -11 2023 +10 Last 12m\n Leisure, hospitality +38 2023 +5 Last 12m\n Construction +18 2023 +9 Last 12m\n Financial +2 2023 -9 Last 12m\n Information -11 2023 -10 Last 12m\n\n \n One month pulls this down. Federal payrolls fell about 166K in October\n 2025, as staff who took a 2025 offer to resign came off the payroll. Without October, jobs\n outside health care grew 15K a month and all jobs 58K. In the last six months, about 32K\n outside health care.\n\n SourcesBLS payroll survey via FRED. Full sources and method on the last page.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 4 of 11\n\n tigzig.com\n\nSlide 5\n\nTIGZIG\n US JOBS · SEPTEMBER 2026\n 4 OCTOBER 2026\n\n \n Health care added jobs in 23 of the last 24 months. Jobs outside it fell in\n 14 of them.\n\n The same nine sectors, month by month. Each chart has its own scale, to\n show its trend.\n\n Jobs added each month, in thousands, January 2023 to September 2026. The October 2025\n drop is the federal one-off on the page before. In February 2026 a strike kept more than\n 30,000 health care workers off payrolls.\n\n All jobs\n Health care, social\n All except health, social\n Government\n Professional, business\n Leisure, hospitality\n Construction\n Financial\n Information\n\n SourcesBLS payroll survey via FRED. Full sources and method on the last page.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 5 of 11\n\n tigzig.com\n\nSlide 6\n\nTIGZIG\n US JOBS · SEPTEMBER 2026\n 4 OCTOBER 2026\n\n \n Unemployment is higher than in 2023 and people stay out of work longer.\n This year unemployment eased because fewer people are in the labour force.\n\n Unemployment rate per cent Jan 2023 Sep 2026 3.5 4.2\n Broad rate, U-6 adds involuntary part-time Jan 2023 Sep 2026 6.7 7.6\n Laid off for good per cent of the labour force Jan 2023 Sep 2026 0.78 1.03\n Out of work 15 weeks+ per cent of the labour force Jan 2023 Sep 2026 1.2 1.8\n Out of work 6 months+ per cent of the labour force Jan 2023 Sep 2026 0.66 1.14\n Median time out of work weeks Jan 2023 Sep 2026 9.6 11.5\n\n In September the unemployment rate rose from 4.14 to 4.18 per cent, within\n the survey's margin of error. More people joined the labour force and most of them found\n work. Since January, participation fell from 62.1 to 61.8 per\n cent and the share of adults in work from 59.4 to 59.2. Both comparisons start after the\n annual population update in January, so they are like for like.\n\n SourcesBLS household survey via FRED. Full sources and method on the last page.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 6 of 11\n\n tigzig.com\n\nSlide 7\n\nTIGZIG\n US JOBS · SEPTEMBER 2026\n 4 OCTOBER 2026\n\n \n Hiring has slowed since 2023 and workers quit less. Layoffs are low, but\n finding work takes longer.\n\n Job openings per person looking Jan 2023 Aug 2026 1.79 1.01\n Hiring rate per cent of employment Jan 2023 Aug 2026 4.1 3.3\n Quits rate per cent of employment Jan 2023 Aug 2026 2.5 1.9\n Layoffs rate per cent of employment Jan 2023 Aug 2026 1.1 1.0\n\n If layoffs are this low, why is \"laid off for good\" on\n page six higher than in 2023? Layoffs count people let go in a month, and that\n stays low. Laid off for good counts people who lost a job for good and are still out of work.\n Few are let go, but those who are take longer to find a new job, so the count built up.\n It peaked early in 2026 and at 1.03 per cent of the labour force is still above the 0.78 of\n January 2023.\n\n The news is full of layoffs. Why is the rate low? About 1.6 million\n people were laid off or let go in August. That happens every month: in 2018 and 2019 it was\n 1.8 million a month. Headlines report the big announcements. As a share of all jobs,\n layoffs are below where they were before the pandemic. These figures come from JOLTS,\n which is published a month later.\n\n SourcesBLS JOLTS via FRED. Full sources and method on the last page.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 7 of 11\n\n tigzig.com\n\nSlide 8\n\nTIGZIG\n US JOBS · SEPTEMBER 2026\n 4 OCTOBER 2026\n\n \n Fewer people are being let go than a year ago, so fewer claim benefit. More\n of those who claim now reach the end of it than in 2023.\n\n New jobless claims thousands a week Jan 2023 Sep 2026 204 200\n Continuing claims thousands, monthly average Jan 2023 Sep 2026 1,604 1,710\n Reach end of benefit per cent of claimants Jan 2023 Aug 2026 31.9 39.2\n\n \n Unemployment is higher than in 2023. Why are new claims not? New claims\n count people let go each week, and that has not risen. Those let go stay out of work longer,\n so the number of unemployed who lost a job is up 24 per cent since January 2023 while new\n claims are flat.\n\n Why are continuing claims lower than a year ago? A new claim becomes a\n continuing claim, and fewer people are starting one: first payments in the three months to\n August were down 7 per cent on a year earlier. The share reaching the end of their benefit is\n about the same as a year ago, 39.2 per cent against 39.8, and well above the 31.9 of January\nSourcesDepartment of Labor and BLS, via FRED. Full sources and method on the last page.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 8 of 11\n\n tigzig.com\n\nSlide 9\n\nTIGZIG\n US JOBS · SEPTEMBER 2026\n 4 OCTOBER 2026\n\n \n Hourly pay beat prices from May 2023 to March 2026. Since April it has been\n behind.\n\n Hourly pay growth on a year, per cent Jan 2023 Aug 2026 4.5 3.1\n Prices, CPI growth on a year, per cent Jan 2023 Aug 2026 6.3 3.4\n Energy prices growth on a year, per cent Jan 2023 Aug 2026 0 8.3 16.0\n Hourly pay minus prices points Jan 2023 Aug 2026 0 -1.83 -0.24\n Pay minus core prices points, no food or energy Jan 2023 Aug 2026 0 -1.05 0.67\n Weekly pay minus prices points, hours included Jan 2023 Aug 2026 0 -1.53 0.36\n\n \n Why it moved. In January 2023, after the 2021-22 inflation surge,\n prices rose 6.3% a year and pay 4.5%. By June 2023 prices had slowed to 3.1% and pay\n at 4.7% was well ahead. In March 2026 energy prices jumped to 12.6% a year. By April\n prices rose 3.8%, above pay at 3.6%.\n\n SourcesBLS, via FRED. Full sources and method on the last page.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 9 of 11\n\n tigzig.com\n\nSlide 10\n\nTIGZIG\n US JOBS · SEPTEMBER 2026\n 4 OCTOBER 2026\n\n \n Spending per person is growing faster than income, so people are saving\n less of their income.\n\n Income per person growth on a year, after prices Jan 2025 Aug 2026 0 0.8 1.1\n Spending per person growth on a year, after prices Jan 2025 Aug 2026 2.6 2.3\n Saving rate per cent of after-tax income Jan 2025 Aug 2026 5.7 4.1\n\n The dotted line is growth over the last 12 months against the 12 months\n before.\n\n Saving rate since 1959 per cent of after-tax income, yearly average 1959 2026 13.5 in 1971 2.3 in 2005 7.9 in 2012 15.1 in 2020 10.3 4.7\n\n Over the last 12 months, spending per person grew 1.9 per cent and\n income per person 1.1 per cent. Those are the dotted lines.\n\nSaving averaged 8.5 per cent of after-tax income from 1959 to 2019. It is\n 4.7 in 2026 so far.\n\n SourcesBEA, revised 30 September 2026, via FRED. Full sources and method on the last page.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 10 of 11\n\n tigzig.com\n\nSlide 11\n\nTIGZIG\n US JOBS · SEPTEMBER 2026\n 4 OCTOBER 2026\n\n \n\nSources and method\n\n Updates my analysis of 24 September 2026:\n On the headline numbers\n the US job market looks fine. The story changes when you dig deeper.\n\n Source\nWhat we use\nSeries\n\nBLS payroll survey\nJobs by month and sector, hourly and weekly pay, hours\nPAYEMS, CES0500000003 and ten more\n\nALFRED\nEach month's jobs number as first reported\nPAYEMS, every vintage\n\nBLS household survey\nUnemployment, U-6, time out of work, laid off for good, participation\nUNRATE, U6RATE, UEMPMED, LNS13025703, CIVPART, EMRATIO and seven more\n\nBLS JOLTS\nJob openings, hires, quits, layoffs\nJTSJOL, JTSHIR, JTSQUR, JTSLDR, JTSLDL\n\nDepartment of Labor weekly claims\nNew and continuing claims, weekly figures averaged by month\nICSA, CCSA\n\nDepartment of Labor ETA 5159\nShare of claimants who reach the end of their benefit, first payments\nETA 5159, c51 and c56\n\nBLS consumer prices\nPrices, core and energy\nCPIAUCSL, CPILFESL, CPIENGSL\n\nBEA\nSaving rate, income and spending per person\nPSAVERT, A229RX0, PCEC96, POPTHM\n\nReuters, RBC\nBreak-even job growth, page 2\n2 October 2026\n\n Unrounded unemployment rate. The unemployed over the labour force, before BLS rounds it.\nOpenings per person looking. JOLTS openings over the unemployed, same month.\nPay minus prices. Yearly pay growth less yearly price growth, in points.\nReach the end of benefit. Final payments over the twelve months to date, divided by first payments over the twelve months ending six months earlier.\nOctober 2025. The household survey and CPI have no reading for it, because of the shutdown.\n\n Data. FRED and ALFRED pulled 4 October 2026; DOL ETA 5159 pulled 2 October 2026.\n\n Use of AI. Parts of this work were prepared with AI and steered by me. I\n reviewed, edited and approved every page. Every data series is pulled from the sources above. The analysis and conclusions are mine.\n\n Amar Harolikar · Decision Sciences & Applied AI\n\n 11 of 11\n\n tigzig.com\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QDesk - Quant Report Desk\n\nExplore: Analysis hub, TREMOR API docs, The deck as a PDF\n\nMore posts: On the Headline Numbers the US Job Market Looks Fine. The Story Changes When You Dig Deeper., Bond Markets Have Been Warning for Months. Now They Have Started Acting., US Diesel Is Above $6 a Gallon for the First Time, and the Cost Is Already Moving Through the Economy, Credit Union Q2 2026: Losses Above Pre-Crisis, and Delinquency Above Pre-COVID in Every Segment, A Correction in AI Valuations Is Looking More Likely, and That View Now Comes From a Central Bank, an Asset Manager and a Bond Investor."
},
{
"slug": "us-lending-data-q1-2026-tigzig-api-jun2026",
"title": "Q1 2026 US Lending Data Live on Tigzig. NCUA + FDIC + NY Fed in One API. Two of Three Not on FRED. AI-Agent Ready.",
"tags": [
"portfolio-quants"
],
"content": "Q1 2026 US Lending Data Live on Tigzig. NCUA + FDIC + NY Fed in One API. Two of Three Not on FRED. AI-Agent Ready.\n\nPublished: 2026-06-12\n\nQ1 2026 update of US lending aggregates is now live on Tigzig - loans, delinquency, charge-offs by segment back to 2002, ready for AI agents (Claude Cowork, Claude Code, Codex, Cursor) and for the human web UI on TREMOR. Three datasets: credit unions (NCUA), banks (FDIC), consumer (NY Fed) - two of those three are not on the FRED API; all three are on the Tigzig API with no auth, free. Plus 100+ macro indicators (US and India) to overlay on the data. All sourced from the regulators' own releases and validated before publishing - methodology and validation are on the site. AI-agent ready surface includes: charts and custom cuts on the web, an open API, an MCP server for Claude / Codex / any agent, and full-database downloads in Parquet, SQLite, DuckDB and CSV. This is the same data behind the recent US credit-union red flag deck (card charge-offs past their GFC peak 10 straight quarters) and the consumer credit stress piece. Point your AI at tigzig.com and it connects to the data, explores, and runs the analysis as you direct.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/us-lending-data-q1-2026-tigzig-api-jun2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/us-lending-data-q1-2026-tigzig-api-jun2026 with header Accept: text/markdown\n\nTags\nportfolio-quants\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: Quant Apps hub, Analysis deck hub, API and MCP catalog\n\nMore posts: Red Flag: US Credit Unions. A $1.73T System at GFC-Era Losses. Card Charge-Offs 10 Straight Quarters Past the Crisis Peak., Auto Losses Past the GFC Peak. Cards Almost There. Credit Unions Already at GFC Loss Rates. NY Fed + NCUA Data Now on Tigzig., Releasing US-CREDIT. For Analysts Tracking US Banking. Interactive Tool Covering FDIC SDI and NY Fed CCP., Tigzig Tremor v2 Is Live. 330+ Macro + Credit Indicators Behind an MCP. 83% Lighter Startup. Just Ask Your AI Agent."
},
{
"slug": "us-life-insurance-red-flag-private-credit-offshore",
"title": "Red Flag in US Life Insurance. $2.4T (7.5% of GDP) in Private Credit and Offshore Reinsurance.",
"tags": [
"portfolio-quants"
],
"content": "Red Flag in US Life Insurance. $2.4T (7.5% of GDP) in Private Credit and Offshore Reinsurance.\n\nPublished: 2026-05-08\n\nUS Life Insurers have $2.4T (7.5% of US GDP) in private credit and offshore reinsurance. $1.2T in Fed Unidentified Misc Assets, $1.2T offshore reinsurance memo. Both nearly doubled in 5 years. PE-affiliated loop - same firms own insurers, credit funds and offshore reinsurers. Doom loop risk if retirees surrender annuities en masse.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/us-life-insurance-red-flag-private-credit-offshore\nMarkdown of the HTML page: send GET https://tigzig.com/post/us-life-insurance-red-flag-private-credit-offshore with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/USINSURANCEREDFLAGMAY2026.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nUS Life Insurers now have $1.2 trillion sitting in the Fed's catch-all bucket - private credit, PE, BDCs, alternatives. Another $1.2 trillion of risk is ceded to non-US reinsurers, mostly in Bermuda and Cayman captives, with PE-affiliated structures (Apollo/Athene, KKR/Global Atlantic) the most-cited examples. That's $2.4 trillion* of exposure - roughly 7.5% of US GDP - to a sector now showing visible stress. And a structural conflict at its core: PE owns both sides of the trade.\n\n*$2.4T is gross exposure - $1.2T direct asset holding + $1.2T of risk ceded offshore. For overlap details and methodology, see Explanatory Notes.\n\nThe Thesis\n\nUS Life Insurers' onshore 'Unidentified Misc Assets' jumped from $649B to $1.2T. Offshore reinsurance to Bermuda/Cayman captives also crossed $1.2T. Combined $2.4T sits in the two least-visible places to look for private credit exposure - onshore in a catch-all asset bucket, and offshore via reinsurance ceded to non-US reinsurers.\n\nWhy this is the red flag:\n\n01 - Same Pocket Risk: PE owns both the insurer AND the credit fund it lends to.\n\n02 - Morgan Stanley: 8% default outlook, near Covid.\n\n03 - Banks marking down private credit loans.\n\n04 - Moody's flips BDCs to negative outlook.\n\n05 - BDCs trading at 22% discount to NAV.\n\n06 - Frauds surfacing: Tricolor, First Brands.\n\n07 - US Treasury convening regulators.\n\nLife: GA / Unidentified Misc Assets\n\n| Metric | Q1 '21 | Q4 '25 |\n|---|---|---|\n| Value (US$) | $649B | $1.2T |\n| % of GA total | 10.3% | 15.2% |\n\nLife: GA / Reserve Credit Reinsured to Non-US Reinsurers\n\n| Metric | Q1 '21 | Q4 '25 |\n|---|---|---|\n| Value (US$) | $584B | $1.2T |\n| % of GA total | 9.3% | 15.6% |\n\nSource: tremor.tigzig.com -> US Insurance\n\nFor definitions, overlap details, and methodology, see Explanatory Notes.\n\nWhat is private credit and reinsurance?\n\nBoth are normal business practices. The red flag is the pace of growth, the affiliated loop and stress now visible in the sector.\n\nPrivate Credit\n\nPrivate credit is lending outside the traditional banking system. Non-bank lenders - private credit funds, Business Development Companies (BDCs), direct lending vehicles - originate loans directly to borrowers. The borrowers are mostly middle-market companies (typically below investment grade or unrated), with a growing slice going to consumer finance and asset-backed lending.\n\nLife insurers don't lend directly. They fund the non-bank lenders - through private placements, BDC bonds, CLO tranches, and fund commitments. Some are now buying the lending platforms outright. The chain runs: insurer capital -> non-bank lender -> borrower.\n\nReinsurance\n\nReinsurance is when an insurer transfers risk to another insurer (the reinsurer) to manage liability and free up capital. Standard practice across the industry. Two main types matter here:\nAsset-intensive reinsurance: the original insurer transfers reserves and assets to the reinsurer. The reinsurer takes both the liability and the investment portfolio.\nRisk-only reinsurance: only the mortality or longevity risk transfers. Assets stay on the original insurer's books.\n\nThe growth concern is asset-intensive reinsurance ceded to offshore reinsurers - mostly in Bermuda and the Cayman Islands, often affiliated with private equity sponsors. That's where the $1.2 trillion in the Fed Z.1 \"Reserve Credit Reinsured to Non-US Reinsurers\" series sits.\n\n\"Life insurers have also steadily increased their investments in risky and illiquid assets over the past decade, and their activity has contributed to the expansion of private credit.\"\n\n\"In private credit markets, loan defaults remained at relatively low levels, but the elevated usage of payment-in-kind (PIK) provisions indicates some borrowers may face repayment difficulties.\"\nFederal Reserve Financial Stability Report, May 2026\n\nWhat's just happened.\n\nApril 2026 - March 2026\n\n01 - Apr 7, 2026 - Moody's flips private credit BDC outlook to negative.\n\nThe first negative outlook from Moody's on the sector in over two years. Driven by redemption exodus from non-traded BDCs and elevated leverage in listed peers.\nReuters, Apr 7 2026\n\n02 - Apr 30, 2026 - Listed BDCs trading at 22% discount to book.\n\nMedian price-to-NAV at 0.78x, mean 0.81x. Market is pricing in roughly a 22% overvaluation in stated loan books.\nRaymond James BDC Weekly, Apr 30 2026\n\n03 - Mar 17, 2026 - Morgan Stanley: direct lending defaults to reach 8%, near Covid peak.\n\nMS strategist Joyce Jiang: \"a private credit shakeout on par with Covid losses is coming.\" AI flagged as the catalyst, with ~26% of BDC portfolios exposed to software.\nCNBC / Morgan Stanley, Mar 17 2026\n\n04 - Apr 1, 2026 - US Treasury convenes regulators on private credit and insurance.\n\nBessent: \"I am concerned with watching how this gets to the regulated financial system.\" Sessions covered fund leverage, ratings, offshore reinsurance, and liquidity.\nTreasury sb0430; Insurance Business, Apr 2 2026\n\n05 - Mar 11, 2026 - JPMorgan marks down private credit loans, restricts new lending.\n\nAnd specifically targets software-exposed portfolios. The biggest US lender to the sector pulling back ahead of expected stress.\nCNBC / Bloomberg / FT, Mar 11 2026\n\nWhat's been breaking.\n\nFebruary 2026 - September 2025\n\n06 - Feb 27, 2026 - Top bankers sued by Tricolor noteholders for \"missing giant red flags.\"\n\nNoteholders allege JPMorgan, Barclays, and Fifth Third \"fueled and perpetuated Tricolor's Ponzi-like fraud.\" Some notes trade below 10 cents. Combined disclosed bank losses already exceed $520M.\nBanking Dive / Claims Journal, Mar 2 2026\n\n07 - Sep 28, 2025 - First Brands collapse: $2.3B in off-balance-sheet debt vanished.\n\nDisclosed exposures: Jefferies' Point Bonita $715M, UBS O'Connor $500M+, Onset Financial $1.9B, Katsumi Global $1.75B. Many positions were marked at par weeks before going to zero. CEO resigned; DOJ criminal investigation underway.\nPitchBook / CNBC, Oct 2025\n\n08 - Jun 2025 / Nov 2025 - Moody's: ~1/3 of $6T in US life insurer assets is private credit.\n\nUp to one-third of the $6T held by US life insurers was in private credit at year-end 2024, and 80% of insurers surveyed plan to grow it further. Some firms have over half their fixed income in difficult-to-trade debt.\nInsurance Business / PYMNTS, Nov 2025\n\nThe Affiliated Loop - All in the Same House\n\nPE owns the insurer. PE owns the credit fund. Insurer lends to fund. PE values the loans. Annuity holder bears the risk.\n\nWhat investors and regulators are saying:\n\n01 - The Conflict - CEPR, March 2026\n\n\"The situation is rife with conflicts of interest, most notably PE firms providing valuations for credit funds with Level 3 loans - and asking annuity holders to trust them.\"\ncepr.net\n\n02 - The Affiliated Share - AM Best, via CEPR Mar 2026\n\nRoughly 20% of investments at Athene's US Life Group and KKR's Global Atlantic now go to affiliated funds. Level 3 assets account for ~1/3 of Athene's and Global Atlantic's total assets.\ncepr.net\n\n03 - The Market Signal - FT, April 14 2026\n\nUS insurance is now one of the lowest-performing sectors in the US investment-grade bond index. Industry private credit ~10% of total assets; PE-affiliated insurers (Athene, Global Atlantic) could top 15%. Wellington's Connor Fitzgerald: \"private credit second derivative trade.\"\npymnts.com\n\n04 - The Doom Loop - Marblegate's Andrew Milgram, via Axios Apr 2026\n\nRetirees worry -> surrender annuities -> insurers liquidate private credit -> more distress -> more surrenders. The chain runs on illiquid assets meeting redemption demands.\naxios.com\n\n05 - The Slow Erosion - Oxford Economics' J. Corominas, via Insurance Journal Apr 6 2026\n\n\"Should private credit losses erode insurer solvency, the resulting contagion would not resemble the bank-run dynamics of 2008, but would instead manifest as a slow, grinding erosion of retirement security - harder to detect in real time, and significantly more difficult to reverse.\"\ninsurancejournal.com\n\nThe regulators saw it coming.\n\nApril 2024 - January 2026\n\n09 - Apr 2024 - IMF GFSR Ch. 2: stale valuations, fragile borrowers, hidden leverage.\n\n\"If private credit remains opaque and continues to grow exponentially under limited prudential oversight, these vulnerabilities could become systemic.\"\nIMF GFSR April 2024, Chapter 2\n\n10 - May 2025 - Boston Fed: bank lines to private credit could pose systemic liquidity risk.\n\nBanks fund private lenders through credit lines that can be drawn together if borrowers default in unison. The Fed's own words: \"tail risk may be underappreciated.\"\nBoston Fed CPP 25-8, May 2025\n\n11 - Aug 2025 - NAIC Actuarial Guideline 55: new asset-adequacy testing for offshore reinsurance.\n\nNew asset-adequacy testing requirement for reinsurance counterparties. Practical bite falls on affiliated Bermuda/Cayman structures.\nClifford Chance, NAIC AG 55, Mar 2026\n\n12 - Oct 2025 - IMF GFSR: ~40% of private credit borrowers have negative cash flow.\n\nUp from 25% in 2021. Stress tests: banks with 18% of global assets would breach safe capital in a stagflationary shock.\nIMF Global Financial Stability Report, Oct 2025\n\n13 - Jan 2026 - Bermuda Monetary Authority tightens reinsurance disclosure rules.\n\nTighter asset/liability disclosure effective Jan 2026. New entrants slipped to 58 in 2025 from 63 in 2024.\nRoyal Gazette, Jan 2026\n\nExplanatory Notes.\n\nDefinitions, caveats, and methodology notes.\n\nOn the 'Unidentified Miscellaneous Assets' Bucket\n\nThe Fed Z.1 Financial Accounts use this label as a catch-all for Life insurer General Account assets that don't fit standard categories. Per Fed methodology, the bucket includes:\nPrivate credit funds, direct lending vehicles, BDC holdings\nPrivate equity fund interests, hedge fund investments\nInfrastructure debt, real estate equity (non-mortgage)\nOther Schedule BA-type alternative investments\n\nNot purely private credit, but the closest aggregate proxy at the industry level for alternatives exposure. Growth from $128B (2000) to $1.2T (2025) reflects a clear structural shift to alternatives.\n\nOn 'Reserve Credit Reinsured to Non-US Reinsurers'\n\nThis is a Fed Z.1 reserve credit memo item, not a direct asset or liability line. It measures how much reserve credit US Life insurers are taking against reserves ceded to offshore reinsurers - a measure of risk and reserve transfer, not gross assets.\n\nWhere the backing assets sit depends on the reinsurance structure:\nCoinsurance with full asset transfer: assets move to the offshore reinsurer\nModco / funds-withheld: assets remain at the cedant under trust or withheld arrangements\nYRT / risk-only: only mortality risk transfers; assets stay with the cedant\n\nOn the Overlap Between the Two Buckets\n\nThe onshore $1.2T sits directly on US Life insurer balance sheets. The offshore $1.2T is a reserve credit memo item - the backing assets sit either at the offshore reinsurer or remain at the cedant under modco / funds-withheld structures. Where assets remain at the cedant, partial overlap with the onshore bucket is possible.\n\nRegulators and the industry track the two as separate exposures because the transmission risk differs - direct credit risk on one side, counterparty risk to a less-regulated reinsurer on the other. The $2.4T combined figure is gross headline exposure.\n\nOn the Denominator\n\nShare % shown for both items is computed against General Account financial assets ($7.7T at Q4 2025), not Total Life Financial Assets ($11.2T). GA is the right base because both items are GA-side in Fed Z.1: Unidentified Misc Assets is a GA-only asset line, and Reserve Credit Reinsured is a GA-side memo line (an offset to gross statutory reserves, not a balance-sheet asset).\n\nData Validation\n\nAll Fed Z.1 series in this brief cross-validated against the Federal Reserve's published L.116 Life Insurance Companies table - exact match across all overlapping line items. NAIC statutory framework comparison from ACLI 2025 Life Insurers Fact Book (80th Ed., year-end 2024 data). Methodology and validation tabs available in the Tremor app.\n\nData Source & Tool.\n\nData Source\nSource: Federal Reserve Z.1 Financial Accounts (via FRED)\nSeries: Life: GA / Unidentified Misc Assets\nCoverage: Q1 2000 - Q4 2025 (104 quarters)\nValidation: Cross-checked vs Fed L.116 and ACLI 2025 Fact Book\nPulled via: tremor.tigzig.com / US Insurance\n\nWhat the Tool Does\n73 series across Life and P&C\n104 quarters of data\nLife: GA + SA breakdown\nP&C: full investment allocation\nOn-demand charts, tree-like nav\nFilters / Methodology / Validations\nFull CSV download\n\nNote on the bucket: Fed's catch-all for assets that don't fit standard categories - private credit funds, BDCs, PE / hedge fund interests, infrastructure debt, real estate equity (non-mortgage), and Schedule BA-type alternatives. Aggregate industry data only.\n\nTremor / US Insurance: Life GA holdings sorted by % change. Unidentified Misc Assets row: +75.6% / +4.84pp.\n\nLive Now: tremor.tigzig.com / US Insurance\n\nExternal Sources.\n\nHyperlinks for all 19 source citations in this brief. Both clickable and shown as raw URLs - so anyone can copy-verify.\n\nWeb version with clickable links: tremor.tigzig.com/ins\n\n01 - Apr 7, 2026\nMoody's cuts outlook on US BDCs to 'negative' - Reuters\nreuters.com/business/finance/moodys-cuts-outlook-us-bdcs-negative-redemption-pressure-rising-leverage-2026-04-07/\n\n02 - Apr 30, 2026\nRaymond James BDC Weekly Insight - April 30, 2026\nraymondjames.com/-/media/rj/dotcom/files/corporations-and-institutions/investment-banking/industry-insight/bdc_update.pdf\n\n03 - Mar 17, 2026\nPrivate Credit Shakeout Matching Covid Losses Coming, Morgan Stanley Says - CNBC\ncnbc.com/2026/03/17/private-credit-shakeout-matching-covid-losses-coming-morgan-stanley-says.html\n\n04 - Apr 1, 2026\nTreasury convenes conversations with regulators - US Treasury\nhome.treasury.gov/news/press-releases/sb0430\nUS Treasury convenes regulators over insurance industry's private credit binge - Insurance Business\ninsurancebusinessmag.com/us/news/breaking-news/us-treasury-convenes-regulators-over-insurance-industrys-private-credit-binge-570672.aspx\n\n05 - Mar 11, 2026\nJPMorgan reins in lending to private credit firms - CNBC\ncnbc.com/2026/03/11/jpmorgan-reins-lending-private-credit-marks-down-software-loans.html\n\n06 - Feb 27, 2026\nJPMorgan, Barclays, Fifth Third sued over Tricolor - Banking Dive\nbankingdive.com/news/jpmorgan-barclays-fifth-third-sued-tricolor-fraud/813504/\n\n07 - Sep / Oct 2025\nFirst Brands Ch. 11 tarnishes private credit - PitchBook\npitchbook.com/news/articles/first-brands-ch-11-tarnishes-private-credit-with-broad-brush-amid-retail-push\nFirst Brands' implosion ripping through private credit - CNBC\ncnbc.com/2025/10/10/first-brands-implosion-lenders-scramble-to-contain-the-fallout-.html\n\n08 - Jun / Nov 2025\nInsurance Business mag - Moody's $6T / 1/3 finding\ninsurancebusinessmag.com/us/news/breaking-news/us-treasury-convenes-regulators-over-insurance-industrys-private-credit-binge-570672.aspx\nInsurers Are Buying Up Private Credit Debt - PYMNTS\npymnts.com/insurance/2025/insurers-are-buying-up-private-credit-debt/\n\n09 - Apr 2024\nGlobal Financial Stability Report - April 2024, Ch. 2 - IMF\nimf.org/-/media/files/publications/gfsr/2024/april/english/ch2.pdf\n\n10 - May 2025\nCould Private Credit Pose a Risk to Stability? - Boston Fed CPP 25-8\nbostonfed.org/-/media/Documents/Workingpapers/PDF/2025/CPP-20250521.pdf\n\n11 - Aug 2025\nNAIC's evolving response to private equity in insurance (AG 55) - Clifford Chance\ncliffordchance.com/insights/resources/blogs/insurance-insights/2026/03/the-naics-evolving-response-to-private-equity-in-insurance.html\n\n12 - Oct 2025\nGlobal Financial Stability Report - October 2025 - IMF\nimf.org/en/publications/gfsr/issues/2025/10/14/global-financial-stability-report-october-2025\n\n13 - Jan 2026\nBMA says broader supervisory changes planned for 2026 - Royal Gazette\nroyalgazette.com/international-business/business/article/20260129/bma-says-broader-supervisory-changes-planned-for-2026/\n\nAffiliated Loop sources\n\n01 - Mar 2026\nYou Bet Your Life: Insurance, Private Equity Comes For Your Annuity - CEPR\ncepr.net/publications/you-bet-your-life-insurance-private-equity-comes-for-your-annuity/\n\n02 - Mar 2026\nAM Best ~20% affiliated-fund stat (via CEPR) - CEPR\ncepr.net/publications/you-bet-your-life-insurance-private-equity-comes-for-your-annuity/\n\n03 - Apr 14, 2026\nInsurance Sector's Private Credit Ties Has Investors Concerned (FT via PYMNTS)\npymnts.com/news/investment-tracker/2026/insurance-sectors-private-credit-ties-has-investors-concerned/\n\n04 - Apr 2026\nInsurance, private credit doom-loop risk (Marblegate / Andrew Milgram) - Axios\naxios.com/2026/04/03/insurance-risk-private-credit\n\n05 - Apr 6, 2026\nSlow erosion of retirement security (Oxford Economics / J. Corominas) - Insurance Journal\ninsurancejournal.com/news/international/2026/04/06/864565.htm\n\nMethodology / Primer sources\n\n01 - May 2026\nFinancial Stability Report - Federal Reserve, May 2026\nfederalreserve.gov/publications/files/financial-stability-report-20260508.pdf\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, TREMOR Macro Indicators, Quants Agent\n\nExplore: Private Credit & NDFI Tracker, Analysis hub, Quant apps hub\n\nMore posts: US Insurance Assets Tracker - 25 Years of Data, 73 Series, Life & P&C, Private Credit and BDC Stress. Discounts at 13-Year Lows. Withdrawals Past the Gate. Regulators Now Arriving., Private Credit Default Rate Hits a Record 6.0% (2026): A Market for Lemons? Banks and Insurers Deep In It., The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top., Bank Loss Rates Look Calm. The Stuff Underneath - Consumer, Private Credit, Non-Banks - Says Otherwise."
},
{
"slug": "us-unemployment-long-term-hidden-jun2026",
"title": "US Unemployment Is 4.3%. The Stuff Underneath Says Otherwise. 2M Jobless Over Six Months, Up 3 Straight Years.",
"tags": [
"portfolio-quants"
],
"content": "US Unemployment Is 4.3%. The Stuff Underneath Says Otherwise. 2M Jobless Over Six Months, Up 3 Straight Years.\n\nPublished: 2026-06-11\n\nUS unemployment is 4.3% and the headline curve looks like it is calming - eased from 4.5% in Nov, held at 4.3% since March. The data underneath says otherwise. Long-term joblessness (27+ weeks) has nearly doubled since 2023 to 2.0 million people, with three straight years up and no give-back. When long-term pulls away from the headline like this, that is the 2008 pattern - into 2008 the long-term line towered about 6x its pre-crisis level while U-3 only doubled. Covid never had this shape; rehiring was too fast. Today's rise has the slow-grind 2008 shape, with no recession to blame. Composition: the long-term share fell to 17.8% after Covid (Feb 2023) and is back to 27.5% now (GFC extreme was 45.5%) - more than one in four of the unemployed are now past the line that scars. Structural test: measured against the whole workforce, long-term joblessness has stepped up once in 25 years - 0.53% to 0.81% in the dot-com era. This cycle is at 1.18% in May and still rising. Is AI doing to the jobless floor what dot-com did? Too early to call - the test comes when this cycle finally turns. Fits the picture from last week's note: consumer sentiment at a 73-year low, valuations near 145-year highs. 5-slide deck on BLS / FRED data through May 2026.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/us-unemployment-long-term-hidden-jun2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/us-unemployment-long-term-hidden-jun2026 with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/USUNEMPLOYMENTLONGTERMJUN2026.pdf\n\nTags\nportfolio-quants\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nSlide 1 - The 4.3% is hiding a problem\n\nUS unemployment · May 2026 data\n\nThat 4.3% is the US unemployment rate. The headline curve looks like it is stabilising - it eased from 4.5% in November and has held at 4.3% since March. Underneath, long-term joblessness - people out of work over half a year - has nearly doubled since 2023: 2.0 million people, and still climbing.\n\nChart: long-term unemployment rate vs headline U-3, both rebased to March 2023 = 100 (the long-term rate's low). Long-term = jobless 27+ weeks as a share of the workforce. Monthly BLS data through May 2026.\n\nNo respite in the data: three straight years up - the long-term rate has not given back a single year.\n\nSlide 2 - We have seen this shape before\n\nAngle 1: the crisis signature\n\nChart: headline U-3 (navy) and the long-term rate (terracotta), both rebased to January 2006 = 100. Shaded bands = recessions.\n\nInto 2008 the long-term line did not just rise - it towered over the headline, about 6x its pre-crisis level while U-3 only doubled. That divergence is the signature of a market where people stop getting re-hired.\n\nCovid was the exception: millions did cross 27 weeks, but rehiring was so fast the surge unwound within about a year. Today's rise has the slow-grind 2008 shape.\n\nThe gap is opening again - with no recession to blame. That is the red flag.\n\nSlide 3 - The mix is moving to the part that scars\n\nAngle 2: composition\n\nChart: every unemployed person, split by how long they have been jobless. Each month adds to 100%. The terracotta band (27+ weeks) is the story.\n\nWhen the terracotta band widens, it means the people who lost their jobs months ago are still searching - re-hiring has stalled. That is exactly what it has been doing since 2023.\n\nThe long-term share fell to 17.8% after Covid (Feb 2023). It is back to 27.5% now. The GFC extreme was 45.5%.\n\nShort-term unemployment is healthy churn. Long-term is the part that scars - skills fade, employers screen it out. More than one in four of the unemployed are now past that line.\n\nSlide 4 - Is AI the next dot-com?\n\nAngle 3: the structural question\n\nChart: the long-term rate (grey line) with its average in each calm stretch between crises (terracotta bars). One step up, then a 15-year plateau.\n\nMeasured against the whole workforce, long-term joblessness has stepped up once in 25 years: from 0.53% to 0.81% in the dot-com era, when work moved to the web. The GFC and Covid spiked and fell back - cyclical episodes, both of them.\n\nThis cycle has not played out: 1.18% in May and still rising - it has not topped out, let alone settled.\n\nToo early to call. The test comes when this cycle finally turns: does the rate settle back near 0.8%, or on a new, higher floor? That is the chart to watch.\n\nSlide 5 - Previous analysis, data tools, sources & method\n\nPrevious analysis - this series\nMarkets at a record while the economy hurts - peak valuations, record-low consumer sentiment. 8 Jun 2026.\nIrrational exuberance: S&P at 7,400 - the last two times this setup held, the index halved. 17 May 2026.\nAre we headed for stagflation-lite? - the Fed caught between inflation and weak jobs. 28 Mar 2026.\n\nThe data tools\nU-1 (15+ weeks), U-3 and U-6 are live on tigzig.com -> Tremor - a macro early-warning dashboard, 100+ indicators with validation built in. The 27+ duration series came via FRED.\nAgent-first portal with MCP servers and free APIs: agents.tigzig.com - point your AI at it.\n\nSources & method\nBLS household survey, via FRED: U-3 UNRATE · duration UEMPLT5 UEMP5TO14 UEMP15T26 UEMP27OV (27+ wks = \"long-term\"). Pulled 10 June 2026; cross-checked to the BLS Employment Situation.\n27+ share = long-term / total unemployed. 27+ rate = long-term / workforce.\nBLS seasonally adjusts each duration series independently, so the buckets sum to total unemployment with a small residual. The survey samples ~60,000 homes, so single months are noisy. October 2025 has no reading (government shutdown).\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent, QRep - Security Reports\n\nExplore: Quant Apps hub, Analysis hub, MCP servers and APIs\n\nMore posts: Index at a Record. Peak Valuation. Record-Low Sentiment. Last Two Times This Setup Held - 2000 and 2007 - the Market Halved., S&P at 7,400. Irrational Exuberance. The Last Two Times This Setup Held, the Index Halved., Are We Headed for Stagflation-Lite, S&P at All-Time High. Under the Hood, Things Are Cracking. Just Like October 2007., Auto Losses Past the GFC Peak. Cards Almost There. Credit Unions Already at GFC Loss Rates. NY Fed + NCUA Data Now on Tigzig."
},
{
"slug": "us-valuations-record-sentiment-low-jun2026",
"title": "Index at a Record. Peak Valuation. Record-Low Sentiment. Last Two Times This Setup Held - 2000 and 2007 - the Market Halved.",
"tags": [
"portfolio-quants"
],
"content": "Index at a Record. Peak Valuation. Record-Low Sentiment. Last Two Times This Setup Held - 2000 and 2007 - the Market Halved.\n\nPublished: 2026-06-08\n\nIndex at a record. Peak valuation. Lowest sentiments. And the economy cracking. Last two times the market was priced up here - 2000 and 2007 - it halved. Shiller CAPE near 41 P/E, the highest in 145 years apart from the 1999-2000 dot-com bubble (44), above the 2007 peak (28). University of Michigan consumer sentiment at 44.8, the lowest in 73 years of data - below the 2022 low (50) and every recession since 1952. In dot-com it was all about page views with no line of sight on revenues; in 2026 it is all about compute and revenues with no line of sight on cash profitability. Companion to the prior 'Irrational Exuberance' note when the S&P was at 7,400. Consumer delinquencies high, unemployment high, producer prices the hottest in 25 years (excl Covid), oil and gasoline high, consumer stretched. Seven stocks doing most of the lifting in the index. Dimon called it too much exuberance. Buffett sitting on the biggest cash pile in Berkshire's history. Drawdown ground-truth from tigzig Market Tools: S&P 500 deepest -56.8% beginning Oct 2007, over five years to climb back to even, 58% of its time spent in the seven worst drawdowns. Not investment advice.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/us-valuations-record-sentiment-low-jun2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/us-valuations-record-sentiment-low-jun2026 with header Accept: text/markdown\n\nTags\nportfolio-quants\n\nNote: this is a short text-only post (release announcement / brief commentary). The full body lives at the HTML page URL above - there is no deck source or PDF transcript for this one. The image referenced in the body (if any) is a generic illustration, not analytical content.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, QRep - Security Reports, Quants Agent\n\nExplore: Market Tools, Analysis archive, Data APIs and MCP\n\nMore posts: S&P at 7,400. Irrational Exuberance. The Last Two Times This Setup Held, the Index Halved., S&P at All-Time High. Under the Hood, Things Are Cracking. Just Like October 2007., Three Red Flags Hit Together. 30Y at 5.14%, PPI at 5.99%, Brent Above $100. Setup Worse Than 2000 or 2007., The Convergence Risk. Credit Stress Already Past 2007 Pre-Crisis. Five Aggravators Stacking On Top."
},
{
"slug": "vibe_coding_andrej_karpahty_mito_ai",
"title": "Vibe coding (Andrej Karpathy) in Jupyter with Mito-AI — the Cursor for data scientists. My top 8 Tips",
"tags": [
"ai-coders"
],
"content": "Vibe coding (Andrej Karpathy) in Jupyter with Mito-AI - the Cursor for data scientists. My top 8 Tips\n\nMito-AI: In the same league as Cursor, Flowise, Lovable, n8n - a powerful and well designed tool. A few weeks of vibe coding with Mito-AI in Jupyter, and I'm hooked.\n\nEfficiency Boost\n\nSmooth UI/UX and fast responses. No more copy-pasting code or tabbing through windows - used to break the flow. Cursor like experience. Still a Colab fan, but now defaulting to Mito-AI.\n\nVibe Coding - Andrej Karpathy on X. Excerpts:\n\n\"There's a new kind of coding I call 'vibe coding'… I barely touch the keyboard, just talk to Composer with SuperWhisper… I accept all suggestions… Copy-paste errors, usually that fixes it… I just see stuff, say stuff, run stuff, and it mostly works.\"\n\nFull post here : Andrej Karpathy on X: \"There's a new kind of coding I call \"vibe coding\"\"\n\nMy top tips for using Mito-AI Copilot\nUse vibe (voice) coding - Win + H (Cmd + H for Macs). If it's not working, go to Settings → Voice Activation Privacy → Turn on access.\nOne chunk at a time - avoid having it do too many things in one prompt. Much faster overall.\nNo need to copy-paste sample data - Mito-AI auto-detects dataframe schemas.\nRefer to previous messages - has full conversation history. e.g: \"You know you did that date extraction from that zip file name… do the same thing for this CSV.\"\nInstall with or without Mitosheet - pip install mito-ai mitosheet adds a nice visual spreadsheet view…but works without Mitosheet also. Try both. I like with Mitosheet.\nOpen Source / Pricing - Free tier (500 messages/month). Premium ($20/month) for unlimited completions + premium models + greater functionality. OR use the base version on an open-source basis with your API key - using this now. If usage increases, I'll move to premium.\nSetting API key –\nSession level (PowerShell): command: $env:OPENAIAPIKEY = \"sk-proj…your-open-ai-key\" … then start notebook from SAME terminal (command: jupyter-notebook)\nModeling - analysis - automation: validations critical - Need to ask the right questions, validate results, sometimes check the code. Best part - tools like Mito-AI & Cursor make this super-efficient too.\n\nVideo\n11-min walkthrough\n\nVideo Time Stamps\n00:00 - Start\n00:08 - Activating Voice Mode & PIP Installs\n01:02 - Basic usage - import / transform & Mito AI schema memory\n04:06 - Example - prompting for EDA / Model Build\n06:48 - Example - connecting to database\n08:23 - Updating OPENAIAPIKEY - environment variable - 2 methods\n\n\nRelated\n\nTools: Practice Lab, BRIQ, Quants Agent\n\nExplore: MCP and agents hub, Database AI apps hub, Claude for Analytics hub\n\nMore posts: Google Colab Data Science Agent vs. Mito-AI Jupyter Copilot. How do they compare? When to use which?, Google on a roll — launches DSA — Data Science Agent on Colab. First impression = just brilliant., Andrew Ng is using Claude Code, OpenAI Codex, Gemini CLI."
},
{
"slug": "vigil-brsr-esg-fy2026-oct2026",
"title": "FY 2025-26 BRSR Data Is Live on VIGIL: ESG Disclosures for 663 of the 750 Nifty Total Market Companies",
"tags": [
"vigil"
],
"content": "FY 2025-26 BRSR Data Is Live on VIGIL: ESG Disclosures for 663 of the 750 Nifty Total Market Companies\n\nReleasing the FY 2025-26 BRSR data on VIGIL. ESG disclosures for 663 of the 750 Nifty Total Market companies.\n\nBRSR is the annual sustainability report SEBI requires from the top 1,000 listed companies. This release covers the workforce, pay and complaints sections.\n\nWhat we cover on VIGIL\nWorkforce: headcount split by permanent and contract, employees and workers, women and men, the share of women up to the board and key management, plus turnover and parental leave\nPay: median pay of women and men for employees, workers, key managers and the board, plus minimum wage compliance and wages by town class\nComplaints: sexual harassment under the POSH Act and seven other complaint types, with how many were resolved\n\nQuestions you can take to it\nHow many women are on the board, in key management and on the shop floor, and how does a company compare with its sector?\nWhich companies pay a large share of their workforce at the minimum wage?\nWhere is the gap between women's and men's pay widest?\nHow did a company's numbers move between FY 2024-25 and FY 2025-26?\n\nHow to use it\n\nCharts by index and sector, a Data Explorer table that copies straight into Excel, a side by side compare of up to five companies, an XBRL viewer with every line of the filing, links to each company's XBRL and PDF, and the full dataset as one CSV.\n\nTool, data-download and docs at https://www.tigzig.com/vigil ➜ BRSR\n\nHow the year views work, and how the data is checked\n\nFew quick points\nEach year view reads one report per company. CY is that year and PY is the previous-year column printed in the same report. So a company's PY in the FY 2025-26 view can differ from its CY in the FY 2024-25 view, when it changed last year's figures in the new report.\nCompanies with a December year-end are placed in the year that holds their year-end, so a calendar 2025 report is in the FY 2025-26 view. A few companies have not filed FY 2025-26 yet and stay in the FY 2024-25 view.\nValues are shown as filed. Where a figure looks wrong, such as wages filed in the wrong unit, it is flagged on the chart and listed in the Data Quality tab with the reason, and it is kept out of the medians where it would distort them.\n\nHow it is built and checked\nMethodology tab: the XBRL tags behind all the charts, the definitions and how each figure is calculated\nData Quality tab ➜ Consistency Checks: the checks run on every filing, year by year\nData Quality tab ➜ PDF Validation: XBRL figures checked against the printed PDF report for the Nifty 100\n\nIndia ESG data, FY 2025-26 on VIGIL\n\n\nRelated\n\nTools: VIGIL - India Red Flag Events, TREMOR - Macro Stress Signals, QDesk - Quant Report Desk\n\nExplore: India red flags hub, Analysis hub, What is new\n\nMore posts: India Inc's BRSR Filings - All 656 in One Analytics Workspace. Wage Gap at Top Management: 68%. 1 in 3 Workers on Contract., Releasing BRSR Analytics Tool. For Tracking Indian Corporate ESG, Workforce, Wages, Complaints., The Whole VIGIL India Red Flag Database Is Now Published for Download. One API Call Takes the Lot., Releasing VIGIL Flags. Filter Companies by One or More Signals Instead of Going Company by Company."
},
{
"slug": "vigil-credit-ratings-pledges-insider-trading-india",
"title": "New Tool Release - VIGIL: Credit Ratings, Pledges and Insider Trading for India Markets",
"tags": [
"vigil"
],
"content": "New Tool Release - VIGIL: Credit Ratings, Pledges and Insider Trading for India Markets\n\nNew tool release for India Markets - VIGIL. Latest Credit Ratings, Pledges and Insider Trading in one place.\n\nPull info on any company in a single click. Clean interface. Instant results. No ads.\n\nCredit Ratings\n\nEvery upgrade, downgrade and reaffirmation across all agencies (CRISIL, ICRA, CARE, India Ratings, Acuite, Brickwork, Infomerics). Filter by company, agency, action, instrument type. Data from January 2024.\n\nInsider Trading\n\nAll SEBI PIT disclosures. Who bought, who sold, how much. Filter by company, person category, transaction type. Data from February 2024.\n\nPromoter Pledge\n\nCurrent snapshot of pledged promoter shares across all companies. Spot who has pledged 50%+ of holdings.\n\nCompany Lookup\n\nProbably the most useful part. Type any company name - get all three data points on one page.\n\nData from NSE filings. Updated daily. Free to use. No login required.\n\nComing soon: Alerts straight to your inbox.\nLive app: tigzig.com/vigil\n\n\nRelated\n\nTools: VIGIL - India Red Flag Events, QRep - Security Reports, Quants Agent\n\nExplore: Market tools hub, Analysis hub, API and MCP catalog\n\nMore posts: Evaluating a stock? Pull its VIGIL Report. Free., New feature on VIGIL: Rating Red Flags (India), Is your company's promoter pledging shares to raise money? Are lenders releasing the pledge - or invoking and taking control?"
},
{
"slug": "vigil-data-api-download",
"title": "VIGIL Data API and Download - 315K+ Records, 7 Tables, Updated Daily",
"tags": [
"vigil",
"database-ai"
],
"content": "VIGIL Data API and Download - 315K+ Records, 7 Tables, Updated Daily\n\nVIGIL's full database is now open. Download or API - your choice. Updated daily.\n\n315K+ records across 7 tables - related party transactions, credit ratings, insider, pledges, takeover, encumbrance & surveillance.\nDownload individual tables as CSV, TSV, Parquet, or SQLite\nOr grab the entire database - all 7 tables in a single 13.6 MB SQLite file\nData dictionary - every field documented with descriptions, data types, and derivation notes\nREST API - No auth. No API key. JSON responses with curl, Python, and JavaScript examples on site. 30 req/min rate limit\n\nFull documentation covers data sources, extraction methodology, and validation results for each table.\n\nRelated party transactions pulled from half-yearly XBRL filings (Mar thru Dec 2025 filings) across 728 Nifty Total Market companies - around 250K transactions with counter-party details, relationship types, amounts, and outstanding balances.\n\nAll public data...sourced from NSE filings...now made available in analyzable and queryable format.\n\nFree. No login. No ads.\n\nvigil.tigzig.com/data-api\n\nVIGIL Data API and Download\n\n\nRelated\n\nTools: VIGIL - India Red Flag Events, BRIQ, DATS-4 Database AI Suite\n\nExplore: API and MCP catalog, Database AI apps, Browse all apps\n\nMore posts: Published Full Validation Reports for Every Dataset on VIGIL, Related Party Transactions - Now Live on VIGIL, New Tool Release - VIGIL: Credit Ratings, Pledges and Insider Trading for India Markets, VIGIL - India Red Flag Events Tracker v2 Release"
},
{
"slug": "vigil-encumbrance-events-india",
"title": "Is your company's promoter pledging shares to raise money? Are lenders releasing the pledge - or invoking and taking control?",
"tags": [
"vigil"
],
"content": "Is your company's promoter pledging shares to raise money? Are lenders releasing the pledge - or invoking and taking control?\n\nEncumbrance Events\n\nVIGIL now tracks encumbrance events - every time a promoter pledges, unpledges, or has shares seized by a lender. These are filed under SEBI SAST Reg 31/32, within 7 working days of each event.\n\nThree event types to watch.\nCreation: promoter pledged shares as collateral. Not always a red flag. Could be the company expanding and the promoter putting up personal stake as guarantee. Shows commitment. But could also be distress - you have to dig deeper into where the money is flowing.\nRelease: pledge removed. Usually means loan repaid or refinanced. Generally positive.\nInvocation: the lender seized the shares. By the time this happens, things have probably gone too far. Definite red flag.\n\nThis sits alongside insider trading, takeover filings, credit ratings, pledge snapshots, and surveillance flags on the company page. One search gives you the full picture.\n\n1,400+ encumbrance events tracked. Presets for creations, releases, invocations, high encumbrance. Filter by company or Nifty indices.\n\nUpdated daily. Free. No Ads. No login.\nvigil.tigzig.com\n\n\nRelated\n\nTools: VIGIL - India Red Flag Events, QRep - Security Reports, Quants Agent\n\nExplore: Quant apps hub, Analysis hub, APIs and MCP\n\nMore posts: New Tool Release - VIGIL: Credit Ratings, Pledges and Insider Trading for India Markets, New on VIGIL: SAST Takeover Disclosures (India), Evaluating a stock? Pull its VIGIL Report. Free., Related Party Transactions - Now Live on VIGIL"
},
{
"slug": "vigil-flags-filter-by-signal-sep2026",
"title": "Releasing VIGIL Flags. Filter Companies by One or More Signals Instead of Going Company by Company.",
"tags": [
"vigil"
],
"content": "Releasing VIGIL Flags. Filter Companies by One or More Signals Instead of Going Company by Company.\n\nReleasing VIGIL Flags. Now you can filter companies based on one or more signals rather than going across multiple tabs or one company at a time. Filter conditions are my judgment, covering credit ratings, promoter buying and selling, outsiders crossing 5%, pledges and lenders invoking them, encumbrance, insolvency and the exchange surveillance flags, the ones I use most often to run my own cuts.\n\nTwo options\nany: companies meeting any of the selected flags (acts like an OR condition)\nall: companies meeting all of the selected flags (acts like an AND condition)\n\nMore flags will be added as I build them for my own cuts.\n\nLive tool, for you: tigzig.com/vigil\n\nWant your AI to download or analyze the data? Just point it to api.tigzig.com/vigil/v1/openapi.json\n\nDownload the whole database. If you would like to run more fine tuned analysis, the whole database is available for download, refreshed four times a day in step with the live data.\n\nHow it is built: go to 'Docs'. Validations: go to 'Validation'.\n\nA few nuances\n\nOne thing to know before you read the numbers, because the twelve conditions are not all the same kind of thing.\n\nEight of them are events with a date on them, so a promoter selling down or a rating moving belongs to the week it was filed, and the period filter does what you would expect. The other four are the position a company is in right now, so over half the company pledged, most of the promoter holding tied up, the severe exchange actions and insolvency carry no date of their own and they read the same whatever period you pick.\n\nThat shows up when you narrow the universe. Over the last three months across the Nifty 750 there are 15 companies where an outsider crossed five per cent and 12 where a promoter sold that much, and just one in insolvency, because almost every company sitting in insolvency is outside that index. Widen it to everything and there are 37.\n\nTwo of these look like duplicates and they come from different filings, so it is worth knowing which is which.\n\nPromoter sold down reads the insider trading disclosures, which are company people declaring their own trades, and it triggers on a cut of one percentage point or more in a single market sale. Promoter sold five per cent or more reads the takeover filings instead, which anyone crossing five per cent has to make, and it is a much blunter cut of the same direction.\n\nA company can show one and not the other, so I kept them apart rather than merging them. Steady trimming shows up in the first and large exits in the second.\n\nThe takeover filings also carry something the insider ones never will, which is an outsider building a stake, because those disclosures cover anybody crossing the threshold rather than only the people inside the company. That is the condition with no insider equivalent at all.\n\nAnalytics tool on public data. For informational purposes. Not investment advice.\n\nVIGIL Flags: filter companies by one or more signals"
},
{
"slug": "vigil-full-validation-reports-published",
"title": "Published Full Validation Reports for Every Dataset on VIGIL",
"tags": [
"vigil"
],
"content": "Published Full Validation Reports for Every Dataset on VIGIL\n\nPublished full validation reports for every dataset on VIGIL - India Corporate Signals Tracker. 315,000+ records across 9 categories. Every finding documented publicly.\n\nRan a detailed AI-assisted audit and validation across all data sources. Specifically targeted at areas where I know from experience things tend to break. Covering - Takeovers, pledge, encumbrance, related party transactions, insider transactions, credit ratings and surveillance.\n\nCross-referenced against independent NSE downloads with detailed data quality and error assessment. Pretty much a full rebuild just for validations.\n\ne.g. one major issue found and fixed - pledge symbol mapping was completely empty. 1,388 of 1,509 companies now mapped. Remaining 120 are delisted or suspended.\n\nThen the source data quality stuff. A company puts year as 1925 instead of 2025, another puts it as 2924. Few records contain \"plz check carefully\" - looks like some internal test data that escaped to production. Twelve takeover records show shareholding above 100%. Same company appearing 27 different ways because company name is free text. And bucket loads more..\n\nJust part of the game when you work with data. You need to validate during build....then again in your daily updates...again after major changes..and again on an adhoc basis....and again..and again..till every exception and error has been beaten out of it...then you take a breath...and you start again..\n\nFull reports on every data source: vigil.tigzig.com - Validations\n\nVIGIL Validation Reports\n\n\nRelated\n\nTools: VIGIL - India Red Flag Events, TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Browse all apps, Analysis hub, Database AI apps\n\nMore posts: VIGIL Data API and Download - 315K+ Records, 7 Tables, Updated Daily, Evaluating a stock? Pull its VIGIL Report. Free., VIGIL - India Red Flag Events Tracker v2 Release, Related Party Transactions - Now Live on VIGIL"
},
{
"slug": "vigil-india-red-flag-events-tracker-v2-release",
"title": "VIGIL - India Red Flag Events Tracker v2 Release",
"tags": [
"vigil"
],
"content": "VIGIL - India Red Flag Events Tracker v2 Release\n\nAre your promoters pledging shares? Insiders buying or selling? Credit ratings dropped? Stock under surveillance?\n\nRevamped interface - filters, presets, CSV export, company cross-references.\n\nScan across Nifty 50/100/500 or search any company.\n\nCompany view - everything on one page.\n\nInfo pages - what does this and that mean.\n\nUpdated daily. Fast. Free. No ads.\n\nwww.tigzig.com/vigil\n\nVIGIL v2\n\n\nRelated\n\nTools: VIGIL - India Red Flag Events, QRep - Security Reports, Quants Agent\n\nExplore: Quant apps hub, Analysis hub, All apps\n\nMore posts: New Tool Release - VIGIL: Credit Ratings, Pledges and Insider Trading for India Markets, Evaluating a stock? Pull its VIGIL Report. Free., New feature on VIGIL: Rating Red Flags (India), VIGIL Data API and Download - 315K+ Records, 7 Tables, Updated Daily, Related Party Transactions - Now Live on VIGIL"
},
{
"slug": "vigil-insider-trading-disclosures-aug2026",
"title": "Insider Trading Disclosures for Indian Listed Companies, on Vigil, Updated Four Times a Day.",
"tags": [
"vigil"
],
"content": "Insider Trading Disclosures for Indian Listed Companies, on Vigil, Updated Four Times a Day.\n\nInsider trading disclosures for Indian listed companies, on Vigil, updated four times a day. You can pull up any company and see whether the people inside it have been buying or selling, when they did it, how much, and what it did to their shareholding.\n\nThese disclosures are a legal requirement. Under SEBI's Prohibition of Insider Trading regulations, promoters, the promoter group, directors, key managerial personnel and designated employees have to report their trades once the value crosses a threshold in a quarter, and the exchange publishes what comes in, so each row you see carries a date and a named person against it, along with the holding before and after.\n\nThere are around 38,000 disclosures across about 1,500 companies going back to early 2024, and if you would rather work offline the whole table downloads as CSV, Parquet or SQLite.\n\ntigzig.com ➜ VIGIL ➜ Insider Trading.\n\nFor field definitions and the data quirks ➜ docs ➜ Insider-trading.\n\nIf you want the data rather than the screen\n\nIf you want the data rather than the screen, the whole insider table downloads as CSV, Parquet or SQLite, and there is a query API if you would rather pull it programmatically and build your own tracker or alert on top of it. Both are open, no key and no signup.\n\nThe endpoint list and the field docs are at tigzig.com/apis/vigil.\n\nIf you group this data by role\n\nIf you group this data by role, the person category comes through as free text from the filer, and across the set there are 18 different spellings for what is really about six roles. Promoters, Promoter and Promoter Group all appear separately, Key Managerial Personnel and KMP both show up, and around one in six rows carry no category at all. I publish it as filed rather than recoding it, so you will want to fold the variants yourself before counting.\n\n\nRelated\n\nTools: VIGIL - India Red Flag Events, BRIQ, TREMOR - Macro Stress Signals\n\nExplore: India red flags hub, VIGIL API docs, API and MCP catalog\n\nMore posts: New Tool Release - VIGIL: Credit Ratings, Pledges and Insider Trading for India Markets, 667 Stocks Are Under NSE and SEBI Surveillance. Almost All of Them Sit Outside the Nifty 500., VIGIL Data API and Download - 315K+ Records, 7 Tables, Updated Daily, Evaluating a stock? Pull its VIGIL Report. Free."
},
{
"slug": "vigil-rating-red-flags-india",
"title": "New feature on VIGIL: Rating Red Flags (India)",
"tags": [
"vigil"
],
"content": "New feature on VIGIL: Rating Red Flags (India)\n\nVIGIL Rating Red Flags\n\nVIGIL now tags every credit rating with a red flag classification. Downgrades, defaults, negative outlooks, watchlists, speculative grades - nine categories that tell you which ratings or companies need your attention.\n\nIt looks at three fields from each rating filing - the action, the outlook text, and the credit grade - and classifies each record into one of nine categories:\nDowngrade, Default, Default Grade\nNegative Outlook\nWatch Negative, Watch Developing, Watch In Rating, On Watch\nSpeculative Grade\n\nSo a company rated 'CRISIL D' gets flagged as Default Grade. A 'Reaffirm' with 'Watch Developing' in the outlook gets flagged as Watch Developing. A plain 'Reaffirm' with 'Stable' outlook? Not flagged.\n\nOut of 10,000+ rating records, this helps surface the where some issue might be happening. The red flag filter works on the Credit Ratings page, Nifty 500 grid, and company pages.\n\nTracking all agencies (CRISIL, ICRA, CARE, India Ratings, ACUITE, Infomerics, Brickwork) from Jan 2024 onwards.\n\nResources\nVIGIL App: https://www.tigzig.com/vigil\n\n\nRelated\n\nTools: VIGIL - India Red Flag Events, TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Quant apps hub, Analysis hub, Browse all apps\n\nMore posts: New Tool Release - VIGIL: Credit Ratings, Pledges and Insider Trading for India Markets, VIGIL - India Red Flag Events Tracker v2 Release, Evaluating a stock? Pull its VIGIL Report. Free., Is your company's promoter pledging shares to raise money? Are lenders releasing the pledge - or invoking and taking control?"
},
{
"slug": "vigil-related-party-transactions-aug2026",
"title": "Related Party Transactions for Around 750 Listed Indian Companies Are Now on VIGIL, With the March Half-Yearly Filings In.",
"tags": [
"vigil"
],
"content": "Related Party Transactions for Around 750 Listed Indian Companies Are Now on VIGIL, With the March Half-Yearly Filings In.\n\nRelated party transactions for around 750 listed Indian companies are now on VIGIL, updated to 12 August 2026 and including the March half-yearly filings. Query it on the app or download the full database.\n\nA related party transaction is business a company does with someone connected to it, a subsidiary, a promoter, a director, or a relative of one, and needs to be disclosed every six months.\n\nThe data has about 3,60,000 + rows covering roughly 44,000 counter parties across three half-year periods. You can browse and filter it in the app, query it through the open API without a key, or take the whole table as one file in CSV, Parquet or SQLite.. which range from 9-14 MBs.\n\nHow much a company files depends almost entirely on how many legal entities its group runs. The median company files under 200 rows for the whole period, and nine companies file more than 5,000 each, so the same disclosure rule produces very different volumes.\n\nLive tool (for humans) at: tigzig.com/vigil ➜ Signals ➜ Related Party\n\nHit the docs and validations sections for more details on the data and pre-publish checks carried out.\n\nAnd the 'Data & API' section to download manually yourself .. or have your AI do it via API ... see below\nAPI Docs (for agents): api.tigzig.com/vigil/v1/openapi.json\nAPI Docs (for humans): tigzig.com/apis/vigil\n\nThis is part of the Tigzig Agent Data Surface at tigzig.com/apis, which also carries mutual fund NAVs going back to 2006, macro and credit indicators, and market data. All of it open, no auth, no key and no signup.\n\nHow much each company files\nAcross 744 companies and the full period, the median company files 195 rows. A quarter file under 91, and the bottom 5% file under 27.\nAt the other end, 81 companies file more than 1,000 and nine file more than 5,000 each: Samvardhana Motherson at 6,653, Adani Ports 6,643, HCL Technologies 6,475, Wipro 6,393, Mahindra & Mahindra 5,893, Bharti Airtel 5,874, Reliance 5,662, Kotak Mahindra Bank 5,237 and Adani Green 5,215.\n662 of the 744 file more than 50 rows.\nA company that runs treasury, shared services, manufacturing and distribution through separate legal entities has to disclose every internal transaction between them, so a high count reflects how the group is organised rather than how much business is being done. About 69% of all rows are transactions with group companies.\n\ntigzig.com/vigil\n\nSome data nuances\nTransaction type has a catch-all. About 56% of rows are filed as \"any other transaction\".\nCounter party is a name, not an identifier. There are roughly 44,000 distinct strings and the same entity can appear spelled several ways, so match on normalised names rather than exact ones.\nPeriod end date is stored as text like 31-MAR-2026, so MAX() on it sorts alphabetically and gives you the wrong answer. Parse it first.\nNegative amounts are deliberate. A negative outstanding balance means payable rather than receivable, and there are around 12,000 of them spread evenly across every transaction type.\nAmounts are in rupees and reasonably consistent, the median transaction is about 62 lakh. Still sanity-check the extreme top before you sum anything.\nI publish what the exchange filed, including the values that are wrong. Anything found is listed on the Known Issues page and flagged on the row itself in the app.\nI run multi-level validation, but I am not going to say that they catch everything. Filing data is messy in ways nobody anticipates, and something will slip through, so I keep looking and tightening the validations. If you find something odd, tell me and I will add it.\n\n\nRelated\n\nTools: VIGIL - India Red Flag Events, BRIQ, DATS-4 Database AI Suite\n\nExplore: India red flags hub, VIGIL API docs, API and MCP catalog\n\nMore posts: Related Party Transactions - Now Live on VIGIL, Is a Large Investor Building a Position or Selling Down in Your Stock? Takeover Filings on VIGIL, Refreshed Four Times a Day., Insider Trading Disclosures for Indian Listed Companies, on Vigil, Updated Four Times a Day., VIGIL Data API and Download - 315K+ Records, 7 Tables, Updated Daily"
},
{
"slug": "vigil-sast-takeover-disclosures-india",
"title": "New on VIGIL: SAST Takeover Disclosures (India)",
"tags": [
"vigil",
"security"
],
"content": "New on VIGIL: SAST Takeover Disclosures (India)\n\nVIGIL SAST Takeover Disclosures\n\nWho's buying into the stock you hold? Who's selling out?\n\nUnder SEBI's Takeover Code (Reg 29), every time someone crosses 5% ownership in a listed company, they have to disclose it. Now live on VIGIL.\n\nTwo types of filings: Reg 29(1) is when someone first crosses the 5% threshold. Reg 29(2) is when an existing 5%+ holder changes stake by 2% or more. Both promoters and outsiders have to file.\n\nThis catches stuff insider trading data doesn't - mutual funds accumulating, strategic acquirers building positions, lenders enforcing pledged shares. A promoter quietly selling 6% in the open market is a very different signal from a family transfer. An outsider accumulating 12% could be an activist building a position.\n\nAround 10,000 SAST records from the last 2 years. Leaderboards for largest acquisitions, promoter selling, outsider accumulation, new 5%+ stakes. Filter by company, transaction type, promoter/non-promoter, Nifty indices.\n\nFull explanation of SAST categories in the Docs section.\n\nUpdated daily. Free. No Ads. No login.\n\nResources\nVIGIL App: https://vigil.tigzig.com\n\n\nRelated\n\nTools: VIGIL - India Red Flag Events, QRep - Security Reports, Quants Agent\n\nExplore: Quant Apps hub, Open data APIs and MCP, All live apps\n\nMore posts: Is your company's promoter pledging shares to raise money? Are lenders releasing the pledge - or invoking and taking control?, Related Party Transactions - Now Live on VIGIL, New Tool Release - VIGIL: Credit Ratings, Pledges and Insider Trading for India Markets"
},
{
"slug": "vigil-takeover-sast-filings-aug2026",
"title": "Is a Large Investor Building a Position or Selling Down in Your Stock? Takeover Filings on VIGIL, Refreshed Four Times a Day.",
"tags": [
"vigil"
],
"content": "Is a Large Investor Building a Position or Selling Down in Your Stock? Takeover Filings on VIGIL, Refreshed Four Times a Day.\n\nIs a large investor building a position or selling down in your stock? Updated on tigzig.com VIGIL with the data refreshed four times a day. Cross 5% of a company and you have to tell the exchange, and every 2% move after that as well.\n\nPull up any company and you can see who has been buying into it, or search an acquirer and see everywhere they have been active. Around 3,600 of the filings come from outside the promoter group, which is where the mutual funds, the insurers and the private equity vehicles show up.\n\nAbout 12,000 filings across 1,500 companies going back to January 2024, and the whole table downloads as CSV, Parquet or SQLite if you would rather work offline.\n\ntigzig.com ➜ VIGIL ➜ Takeover.\n\nDetailed explanation of what the filings are, who has to file and when, and details of the fields ➜ Docs\n\nHow the non-promoter filings look\n\nHere's how the non-promoter filings look like. Around 3,600 filings come from outside the promoter group. Over the last twelve months that is roughly 235 large purchases across 174 companies by mutual funds and LIC alone, against about 140 sales across 112 companies.\n\nA point regarding the non-promoter view. Debenture trustees sit in that group too. They hold shares as security against pledged stock rather than as investors, so they are 'non-promoter' but they are not somebody taking a view on the company.\n\nOr just hand it to your AI agent\n\nYou can click around the app yourself, or you can just hand it to your AI agent. No key, no signup, nothing to install. Paste this at Claude, ChatGPT or whatever you use:\n\n``\nRead api.tigzig.com/vigil/v1/openapi.json so you know what this API does. Now grab\nme the whole takeover table. Then tell me who has been building or cutting large\nstakes over the last six months, which companies the mutual funds have been moving\ninto and out of, and anything that looks unusual. After that do the same for\n[YOUR COMPANY].\n``\n\nThere are six other datasets covering corporate filings with red flag potentials. Explore on the tool or download and analyze.\n\nIf you'd like to download it yourself just go to the app ➜ Data & API\n\n\nRelated\n\nTools: VIGIL - India Red Flag Events, BRIQ, TREMOR - Macro Stress Signals\n\nExplore: India red flags hub, VIGIL API docs, API and MCP catalog\n\nMore posts: Insider Trading Disclosures for Indian Listed Companies, on Vigil, Updated Four Times a Day., New on VIGIL: SAST Takeover Disclosures (India), 667 Stocks Are Under NSE and SEBI Surveillance. Almost All of Them Sit Outside the Nifty 500., VIGIL Data API and Download - 315K+ Records, 7 Tables, Updated Daily"
},
{
"slug": "vigil-whole-database-download-sep2026",
"title": "The Whole VIGIL India Red Flag Database Is Now Published for Download. One API Call Takes the Lot.",
"tags": [
"vigil",
"database-ai"
],
"content": "The Whole VIGIL India Red Flag Database Is Now Published for Download. One API Call Takes the Lot.\n\nThe whole VIGIL India red flag database is now published for download. So if you are screening a few hundred companies one API call at a time, take the lot instead. A single API call downloads the whole database. Or download it manually if you would like.\n\nIf you are going to schedule this rather than pull it by hand, read the manifest first. It sits at api.tigzig.com/vigil/v1/downloads/manifest and carries the row counts, the file sizes and a generated_at stamp, so a job can check whether anything has changed before it downloads anything.\n\nThe files answer conditional requests as well. Send If-None-Match with the ETag you got last time and you get back a 304 with no body when nothing has changed. The rebuild runs four times a day, so a job polling more often than that is downloading the same bytes over and over, and that one header removes all of it.\n\nIt is current to 1 September and rebuilt four times a day. Related party transactions are the exception, because those are filed half yearly and run to March, and the next set arrives once the September half closes.\n\nDocs\nFor you: tigzig.com/apis/vigil\nFor your agent: api.tigzig.com/vigil/v1\n\nThe whole VIGIL India red flag database, published for download\n\nWhat is in it\n\nThe downloads and APIs cover related party transactions, insider trading, credit ratings, promoter pledge, encumbrance events, takeover and SAST filings, and the surveillance flags.\n\nAround 2,500 listed companies and about 426,000 rows in all.\n\nOne SQLite file with all seven tables in it comes to roughly 18 MB. If you only want a slice, every table is published on its own as parquet, CSV or TSV, gzipped or zipped, which is 51 files in total.\n\nDetailed docs, the data dictionary, how the data is pulled, the validations carried out and the anomalies detected are all available on the VIGIL site in their respective sections: tigzig.com/vigil\n\nA few quick points on the data\n\nRelated party transactions is the largest by a wide margin, and these are filed on a consolidated basis, so a company's rows cover the listed entity together with its subsidiaries rather than the parent on its own. Some groups file under several hundred entity names, which is worth knowing before you count rows per company and read it as activity.\n\nCredit ratings, insider trading, the takeover and SAST filings and the encumbrance events accumulate over time, so they are event streams you can run a date filter over. Promoter pledge and the surveillance flags are current snapshots instead, replaced on each run, so they tell you where things stand today and carry no history.\n\ntigzig.com/apis/vigil\n\n\nRelated\n\nTools: VIGIL - India Red Flag Events, BRIQ, DATS-4 Database AI Suite\n\nExplore: VIGIL API docs, API and MCP catalog, Database AI hub\n\nMore posts: Pulling Tigzig Data Through the API? There Is a Second Way In. The Whole Database Is Published for Download., Related Party Transactions for Around 750 Listed Indian Companies Are Now on VIGIL, With the March Half-Yearly Filings In., Insider Trading Disclosures for Indian Listed Companies, on Vigil, Updated Four Times a Day., Is a Large Investor Building a Position or Selling Down in Your Stock? Takeover Filings on VIGIL, Refreshed Four Times a Day., VIGIL Data API and Download - 315K+ Records, 7 Tables, Updated Daily"
},
{
"slug": "voice-mode-query-analyze-database-aws-azure-custom-gpt",
"title": "VOICE MODE - Querying & Analyzing Data with Custom GPT AWS - Azure Data Warehouse",
"tags": [
"voice-ai",
"custom-gpt",
"database-ai"
],
"content": "VOICE MODE - Querying & Analyzing Data with Custom GPT AWS - Azure Data Warehouse\n\nSee me talking to Custom GPT in voice mode. At the backend, it is talking to Azure Data Warehouse (MySQL). ... and analyzing profile of campaign responders ..\n\n... am inserting conditional fields, asking questions about data, creating distributions based on calculated fields, and generating charts.\n\nNot in the video, but I have also carried out a bunch of other tasks in voice mode: creating summary tables, merging summary tables back with modeling datasets, appending tables, dropping tables, etc.\n\nApplications\n\nSenior Leadership Voice Dashboards / Adhoc query support / Quick Queries & Charts / Rapid data transformations / Intelligent IVR / Employee queries ... numerous applications ..\n\nThis is Part-2 of my series on connecting to and analyzing data on live data warehouses on AWS & Azure via Custom GPT and LLM Apps.\n\nLink to Part-1\n\n[DW Series -Part 01] Analyze Live Data | AWS–Azure DW | via Custom GPT & LLM Apps\n\nPart-01: Lighthearted Introduction.\n\nUPCOMING EPISODES : COMING NEXT\n\nGPT-LLM Capability Demonstration Videos\n\nData transformations, analysis, charts, table operations, inter-warehouse operations, operating on large data sets, ML model build.....and limitations, caveats & constraints\n\nHow-To Guides\n\nWith Codes / Schemas / GitHub Repos\n\nConnecting to data warehouses, deploying FastAPI Server, GPT action schemas, deploying on external LLM Apps, security issues, LLM cost & options, prototype warehouse setup on AWS & Azure.\n\n👉 With special focus on how to use GPTs to get all this done quickly and efficiently\n\n\nRelated\n\nTools: Voice DB - OpenAI Realtime, Voice DB - ElevenLabs, ChatGPT to Any Database\n\nExplore: Custom GPTs hub, Database AI Apps, MCP hub\n\nMore posts: Analyze Live Data | AWS-Azure DW | via Custom GPT & LLM Apps, GenAI App | LLM Analytics Assistant: Simplifying Data Transformation & Insights. AWS & Azure MySQL DW Example, Realtime voice AI - OpenAI WebRTC Implementation. Live app. Open source."
},
{
"slug": "which-ai-coder-should-you-use-for-xlwings-lite-python-in-excel",
"title": "Which AI Coder should you use for xlwings Lite (Python in Excel)?",
"tags": [
"ai-coders",
"xlwings-lite",
"python-in-excel"
],
"content": "Which AI Coder should you use for xlwings Lite (Python in Excel)?\n\nBeginners: Gemini 2.5 Pro on aistudio.google.com. Free, simple, 1M context. Covered in my guides at xlwings-lite.tigzig.com\n\nHeavy iterative work: Use proper AI coders that handle large codebases with incremental edits - Claude Code, Cursor, VS Code with Copilot, Gemini CLI.\n\nMy default for xlwings Lite: Gemini CLI. Nearly all my xlwings Lite work runs here. Simple setup. Strong free tier, 1M context.\n\nHelp your AI Coder: upload the AI Coder Instruction File I share in my xlwings guides. It is designed to make AI Coder's job easier and output reliable.\n\nWhy not ChatGPT?\n\nIt works fine for quick iterations, but on longer codes it forgets to check the instruction file consistently and the context window fills up too fast.\n\nClaude?\n\nBest AI coder by a wide margin. I use both Claude Code as well as Cursor. But both have been clamping down on message limits - so I toggle between AI Coders depending upon task and messaging quota remaining and typically keep them for more complex pieces.\n\nGemini CLI?\n\nNot Claude-level, but excellent for xlwings Lite, Python FastAPI backends, simpler HTML-JS UI's. With the Instruction File it works out superbly well for xlwings Lite. And the free tier makes it practical. This is my current setup too.\n\nResources\nGemini CLI - How to Install, Setup and Use Tutorial (YouTube)\nxlwings Practice Lab - AI Coder Instructions File, hands-on guides, resources and live xlwings Lite apps\nxlwings lite Official site and docs\n\nAI Coder for xlwings\n\nxlwings Lite official site: lite.xlwings.org\n\n\nRelated\n\nTools: Technical Analysis in Excel, DUCKIT - CSV to DuckDB, Data Importer\n\nExplore: Practice Lab, AI in Excel Hub, MCP Hub\n\nMore posts: The xlwings Lite AI Coder Instruction File - December 2025 Release, Python In Excel - Claude Vs. xlwings Lite? Who Wins?, Python in Excel: Field Guide & Practice Lab for AI-assisted xlwings Lite., AI Coders give you the edge.The 6 Rules I Follow When Working with AI Coders."
},
{
"slug": "working-with-claude-cowork-claude-code-jun2026",
"title": "Claude Cowork and Claude Code Have Quietly Become My Universal Front End. More Than Excel, Jupyter or Colab.",
"tags": [
"ai-coders",
"mcp"
],
"content": "Claude Cowork and Claude Code Have Quietly Become My Universal Front End. More Than Excel, Jupyter or Colab.\n\nPublished: 2026-06-18\n\nFirst-person guide on running almost the whole analysis-and-modeling pipeline from one place. For the author, Claude Cowork (and Claude Code) have quietly become the universal front end - past Excel, Jupyter and Colab - covering exploration, cleaning, ensembles, hyperparameter tuning, validation, and final presentation deck. Each iteration is shared as HTML reviewed in the browser, so a session ends with both a model AND full documentation written as work progressed. Same flow works for database-connected analysis, not just model builds. Excel work goes through Claude in Excel or Cowork; PowerPoint too. Full-stack builds stay on Claude Code. The difficult part was letting go: after years of notebooks and code editors, with the comfort of code you can touch and feel, stepping back to talk into a chat box and review the results took some getting used to. Once you let go, it is way faster and gives much better results, since the human focuses on validations, logic, business and domain pieces - the places where an AI agent still needs guidance. Includes a 23-slide deck that Cowork put together about itself: chat -> coworker shift, desktop + Code teammate, hitting APIs (FRED, Yahoo Finance), reaching backends via MCP, working files in place, cleaning messy spreadsheets, 11.85M-row / 1.6GB analyses in 13 seconds, EDA-as-browser-page iteration loop, gradient-boosting model builds with ROC / gains, A/B significance tests, packaging as HTML/PDF/Word/PPT/Excel, live dashboards that pull fresh data, real URL fetching (not snippets), scheduled briefings, browser/desktop control, cross-session memory, sandbox toolkit (4GB RAM / 10GB disk / pandas / polars / pyarrow / scikit-learn / matplotlib / OpenCV / Pillow), no-code-on-your-machine handoff to Claude Code, and Anthropic's own resources (Cowork product, best-practices guide, the HTML-beats-Markdown post, computer-use docs). Deck self-built in Cowork as HTML then exported to PDF.\n\nWhere to find the full content\nHTML page (full text, image deck, links): https://tigzig.com/post/working-with-claude-cowork-claude-code-jun2026\nMarkdown of the HTML page: send GET https://tigzig.com/post/working-with-claude-cowork-claude-code-jun2026 with header Accept: text/markdown\nPDF deck (full analysis, charts, sources): https://tigzig.com/files/WORKINGWITHCLAUDE_JUN2026.pdf\n\nTags\nai-coders, mcp\n\nFull analysis transcript (extracted from PDF deck)\n\nThis text was extracted from the source PowerPoint deck. Chart visuals (referenced as \"no text content - see PNG at /files/...\") are in the PDF and slide images on the HTML page.\n\nSlide 1 - I'm Claude Cowork\n\nHand me your data and a question, and I'll pull it, clean it, analyze it, build a model, and write the whole thing up as a report or PDF. There's a good deal more I can do, too. Let me show you...\n\nSlide 2 - Chat could only talk. I can act.\n\nA chat assistant answers questions inside a box. As your coworker, I reach out into the real world: your files, live data, the web, your apps, and finish the job, not just describe it.\n\nRead your files · pull live data · run analysis · write reports.\n\nHow: Talk to me the way you already do, and I do the work.\n\nSlide 3 - I sit on your desktop, beside your chat.\n\nI'm a desktop app, not just a browser tab. Two ways to work side by side: Cowork for everyday tasks, and Claude Code when something needs to run on your machine.\n\nHow: Get the Claude desktop app, open Cowork, and start a session.\n\nSlide 4 - Just posted to the FRED API? Already on my desk.\n\nPoint me at an API, and the moment a number lands, I pull it and chart it. US retail sales just posted to the FRED API: a record $763.7B, up 0.9% on the month. With inflation at 4.2%, plenty of that is higher prices, not extra spending.\n\nHow: Point me to the API docs. Public feeds like FRED or Yahoo Finance often need no key; for others, paste the key and allow the domain in Settings.\n\nSlide 5 - I reach your data backends, too.\n\nI can reach your own systems two ways: straight through your API, or via an MCP server. Point me at tigzig.com's Tremor and I pull live data from the back end.\n\nHow: Give me the API docs, or add your MCP server as a connector in Settings. Then just ask.\n\nSlide 6 - I work your files where they live.\n\nGive me access to a folder and I read, edit, rename and organise what's in it, right on your drive. The work stays where you keep it, ready to use.\n\nRead & edit · rename & sort · tidy folders · save final output.\n\nHow: Pick the folders I'm allowed to touch. I'll work inside them.\n\nSlide 7 - Hand me a messy file. Get back a clean report.\n\nA buggy spreadsheet landed in your inbox and you need an answer fast. I clean it, check it, find the story, and hand you a tidy report. You review the numbers, we refine, the analyst still owns the call.\n\nFix broken rows · spot bad data · de-duplicate · report out.\n\nHow: Drop the file in a folder I can see, and tell me what you want from it.\n\nSlide 8 - Big data doesn't go to my head.\n\nAn 11.85-million-row, 1.6GB file? Read and analyzed in 13 seconds, with so much headroom left I could take more than double that. Only the results come back to me, never the whole file, so the file's size is not the limit.\n\nHow: Just point me at the file. Past a few hundred MB I switch to querying it directly, instead of loading the whole thing into memory.\n\nSlide 9 - I don't just fetch. I analyze.\n\nI explore before I conclude, cutting the data every way to see what's really going on, and I show you the results as I go. Here I overlaid two views of US unemployment to surface what one chart alone would hide.\n\nHow: Tell me the question. I'll dig in and show you results as I go.\n\nSlide 10 - I don't stop at charts. I build models.\n\nPoint me at a target and I train on the whole file, then validate it on held-out data.\n\nGradient boosting · scikit-learn · ROC, gains, confusion.\n\nHow: Tell me what to predict. I'll train it, validate it, and show the curves.\n\nSlide 11 - I run the stats, too, and tell you what holds up.\n\nResponse rates, conversion rates, A/B tests, the bread and butter of marketing. I run the test and tell you whether the difference is real, not just noise.\n\nHow: Give me the two groups. I'll run the test and call it.\n\nSlide 12 - I show my work, and we iterate.\n\nNo black box, and no one-shot. I share the exploration first, the missing values, the oddities, the data issues, as a page you can open right in your browser. We fix and refine, then build the model, then tune it together.\n\nEDA & missing values · open it in your browser · we refine together.\n\nHow: Ask me to share the exploration as a page you can open in your browser.\n\nSlide 13 - I package it the way you need to send it.\n\nA clean one-pager or a multi-page report, then out as whatever the moment needs, including a PDF with clickable links.\n\nHow: Just say the format: HTML, PDF, Word, PowerPoint or Excel.\n\nSlide 14 - I'll build you a live dashboard.\n\nNot just a static report, a page you reopen tomorrow and it pulls fresh data each time. A tracker, a metrics page, a queue you keep coming back to.\n\nLive tracker · metrics page · refreshes itself.\n\nHow: Ask for \"a live page I can reopen\". It pulls fresh data each time.\n\nSlide 15 - I read the real web page, not a snippet.\n\nGive me a link and I open the actual web page and fetch the whole thing with my own tools, the real text, not just a short search summary. So I can quote it properly and cut down on guesswork.\n\nOpen the real URL · read the whole page · less guesswork.\n\nHow: Give me the link or the topic. I'll open the page and read it.\n\nSlide 16 - I run on a schedule.\n\nSet it once and walk away. A morning briefing, a weekly data refresh, a Monday status check, I'll run it on time and have the result waiting for you.\n\nDaily briefing · weekly refresh · month-end report.\n\nHow: Say \"every morning at 7\" or \"every Monday\", and I'll set it up.\n\nSlide 17 - I'll drive your browser and desktop.\n\nWhen a tool has no tidy API, I'll use it the way you would, clicking through a website or a desktop app to get the thing done.\n\nHow: Ask me to open a site or app. I'll connect to your browser when you're ready.\n\nSlide 18 - I remember you.\n\nTell me your style once, your fonts, your colours, the way you like reports, and I carry it forward. Point me at a preferences file in your project, or just say \"remember this\".\n\nYour style · a preferences file · across sessions.\n\nHow: Point me to your preferences file, or say \"remember this\", and I apply it next time.\n\nSlide 19 - What's in my toolkit.\n\nA private Python sandbox - roughly 4GB of memory and 10GB of disk to work in. Install almost anything from PyPI - the whole Python data and ML stack is open to me.\nData: pandas, polars and pyarrow, to wrangle files far bigger than memory.\nModels & charts: scikit-learn for machine learning, matplotlib for the visuals.\nImages and more: OpenCV and Pillow for image work, and since almost any package installs, audio and other toolkits are a step away.\n\nThe fine print: the very biggest packages (XGBoost, TensorFlow, up to ~600MB) can run past the current 45-second install window and not finish; most install in seconds.\n\nBeyond the sandbox, I plug into your tools - Gmail, Drive, Slack, Notion, and a growing list of data providers, whatever's been made available. And for your own systems, custom connectors through MCP servers.\n\nSlide 20 - What I can't do (and the handoff).\n\nI can't run code directly on your laptop, and I won't move money or place trades. For code that has to run on your machine, my teammate Claude Code handles it: I write the instructions, it executes them.\n\nNo code on your machine · no trades or transfers · Claude Code, same team.\n\nHow: Claude Code sits right there, in the desktop app or in VS Code. Hand it the note and it runs on your machine.\n\nSlide 21 - How Amar works with me.\n\nAmar runs an AI-agent-first shop at tigzig.com. He points me at data, on Tremor and public APIs, I draft, analyse and build, and Claude Code executes on his machine. I've built him HTML one-pagers, PDF reports and models.\n\nHow: Visit tigzig.com for free data and APIs. Want me to explore it? I can reach the site directly from Cowork.\n\nSlide 22 - Resources\n\nEverything above runs in Claude's Cowork mode. The guides, and a couple of good reads:\nCowork, the product: claude.com/product/cowork\nGetting started with Cowork (best practices): claude.com/blog/best-practices-for-getting-started-with-claude-cowork\nWhy HTML beats Markdown, by Thariq Shihipar (Anthropic): claude.com/blog/using-claude-code-the-unreasonable-effectiveness-of-html\nLet Claude use your computer: support.claude.com\n\ntigzig.com - the analyst's tool shed · AI-agent first · 40+ live apps · 200+ build guides · security checklist · MCP & API servers. Point Claude Code or any AI agent at it, and it picks up the work for you.\n\nSlide 23 - By the way, I made this too\n\nThis whole deck was built in Cowork as HTML, then exported to a PDF. Same tools, same desk.\n\nClaude Cowork, with Amar Harolikar · tigzig.com\n\n\nRelated\n\nTools: DATS-4 Database AI Suite, TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Claude for Analytics, Database AI Apps, MCP servers and agents\n\nMore posts: Claude - The Universal Agent, Claude in Excel + MCP + xlwings Lite + Claude Code: Pull Data, Build a Model, Validate It., Claude for Analytics on Tigzig.com - New Hub: Excel, PowerPoint, Cowork, 7 MCP Servers, 20+ Guides., Boris Cherny: Engineering, Product, Design, Data Science Are Melting Into Archetypes. Seeing the Same., AI Coders give you the edge.The 6 Rules I Follow When Working with AI Coders."
},
{
"slug": "xlwings-lite-data-importer-v2-released",
"title": "xlwings Lite Data Importer v2 Released",
"tags": [
"xlwings-lite",
"python-in-excel",
"duckdb"
],
"content": "xlwings Lite Data Importer v2 Released\n\nxlwings Lite Data Importer\n\nNow import any file type, any size (tested up to 1.2GB) into xlwings Lite with options for raw import, auto conversion to DuckDB, token based vs. shareable link.\n\nv2 removes three major limitations in v1: support for data files only, their forced conversion to DuckDB and downloads via shareable links only.\n\nWhat's New in v2\n\nImport any file as-is - PDFs, images, ZIPs, data files. Process as-is. Or auto-convert data files (CSV, JSON, SQLite) to DuckDB. Your choice.\n\nNew download option added - token-based access (fully private, one-time setup). Dropbox, Google Drive, GitHub. Based on token access pattern from Ruud van der Ham's xlwings_utils package. You can now use shareable links (fast, less secure) or token based (one-time setup, fully secure) - your choice.\n\nUsing the Tool\n\nThe tool code is largish to handle multiple import options, user interactions, error handling, dashboards. Best way to use it - cannibalize it - extract what you need, discard the rest.\n\nTop tip - for large CSV/TXT, convert to Parquet first (just have your AI coder do it, or do it in Jupyter, Colab, or in-browser with Duckit App). Faster load, faster queries.\n\nSetup Guides\n\nCORS proxy & token setup guides: simple explanations of what they are and why we need it. Step-by-step for DIY or give to your AI Coder - includes AI instructions. It will set it up for you and prompt you where human help is required.\n\nResources\nxlwings Lite Data Importer - Live tool\nxlwings_utils by Ruud van der Ham - Token access pattern reference\nBuilt with xlwings Lite by Felix Zumstein\n\nxlwings Lite official site: lite.xlwings.org\n\n\nRelated\n\nTools: Data Importer, DUCKIT - CSV to DuckDB, Data Tools\n\nExplore: xlwings Lite Practice Lab, Claude for Analytics hub, Database AI apps\n\nMore posts: DuckDB Meets Excel: xlwings Lite Data Tools, xlwings_utils: Secure Cloud Access & VBA Bridge, xlwings Lite Local File Access: 8 Patterns You Can Use Today"
},
{
"slug": "xlwings-lite-local-file-access-8-patterns",
"title": "xlwings Lite Local File Access: 8 Patterns You Can Use Today",
"tags": [
"xlwings-lite",
"python-in-excel"
],
"content": "xlwings Lite Local File Access: 8 Patterns You Can Use Today\n\nA few weeks ago, Felix Zumstein released local file system access for xlwings Lite. This is a big deal. Until now, xlwings Lite scripts running in the browser had no way to read or write files on your local drive. You could work with data inside Excel, but you could not save a PDF to your desktop or load a CSV from a folder. That limitation is now gone.\n\nI have been experimenting with this feature and compiled 8 useful patterns into a single demo script that you can run end-to-end. The demo uses ODI cricket data and is structured as a 9-step sequential process - the first step simply fetches the sample data so you have something to work with, and the remaining 8 steps each show a distinct local file pattern. You can of course use your own data instead of the demo dataset.\n\nOne-Time Setup: Mount Your Local Folder\n\nBefore running anything, you need to mount a local folder. This is the only setup required.\nOpen xlwings Lite and click the hamburger menu (three lines) in the top-left of the xlwings Lite task pane.\nClick \"Files\".\nGo to the \"Local Folders\" tab.\nClick \"Add Folder\" and select the folder on your computer you want to use (for example, a folder on your Desktop).\nIn the \"Path in Python\" field, type: xltest22\nClick the tick mark button to confirm.\nRestart or reload xlwings Lite for the mount to take effect.\n\nYour local folder is now accessible at /xltest22 inside the Python environment. The script reads from and writes to this path.\n\nGetting the Data In\n\nThe first step downloads a zip file containing about 6,000 individual cricket CSV files from a GitHub release. Since GitHub release URLs are not CORS-enabled for browser access, the download is proxied through a Cloudflare Worker. The file is saved directly to your mounted local drive. You can skip this step entirely if you already have your own data on the mounted drive.\n\nThe 8 Patterns\n\nPattern 1: Unzip and Process Files on Local Drive\nThe script opens the zip file from the local drive, extracts about 3,000 ball-by-ball CSV files (the zip contains 6,000 files total - half ball-by-ball, half match data), and concatenates them into a single 192 MB CSV. The output is written back to the local drive. Uses Python's built-in zipfile and csv modules to read and write files directly on the mounted drive.\n\nPattern 2: Convert CSV to Parquet\nThe combined CSV is read using DuckDB and converted to Parquet format, bringing it down from 192 MB to under 5 MB. The Parquet is stored in browser memory as a working file. Uses DuckDB's readcsvauto and COPY TO commands for format conversion.\n\nPattern 3: Run Analytics and Create Charts in Excel\nDuckDB queries the Parquet data to produce Top 15 Batsmen and Top 15 Bowlers summary tables. Matplotlib generates charts for both. Results are written to Excel sheets with formatted tables and embedded chart images. Uses DuckDB for SQL analytics, Matplotlib for charts, and xlwings to write formatted output back to Excel.\n\nPattern 4: Generate HTML Report to Local Drive\nCreates a complete HTML report with styled tables and chart images, then writes the HTML file along with chart PNG files to the mounted local drive. You can open the report in your browser directly from your desktop. Uses Pandas to_html for table rendering and standard file I/O for saving.\n\nPattern 5: Generate PDF Report to Local Drive\nCreates a 4-page landscape PDF with charts and data tables, saved directly to the local drive. ReportLab is installed on-the-fly via micropip at runtime - no pre-installation needed. Uses ReportLab's platypus layout engine for tables, images, and page formatting.\n\nPattern 6: Save Parquet File to Local Drive\nCopies the Parquet file from browser memory to the mounted local drive, persisting it for use outside xlwings Lite. Uses standard Python file I/O (open, read, write) to transfer the file from browser temp storage to the local filesystem.\n\nPattern 7: Send Email with Local File Attachments\nReads the PDF and HTML files from the local drive, base64-encodes them, and sends them as email attachments via the Brevo transactional email API. The recipient is read from an Excel cell, the API key from an environment variable. Uses pyfetch to call the REST API with file attachments.\n\nPattern 8: Generate PowerPoint to Local Drive\nCreates a 2-slide widescreen PPTX with charts and styled data tables. python-pptx is installed on-the-fly via micropip. Slide 1 has a batting chart and stats table, Slide 2 has bowling stats and chart. The file is saved to the local drive. Uses python-pptx for slide layout, tables, images, and text formatting.\n\nWhat This Means\n\nThe local file mount changes what xlwings Lite can do. Before this, everything was confined to the browser and the Excel workbook. Now you can download files from the internet to your drive, process data using your local filesystem as working storage, and generate output files - PDFs, HTML reports, PowerPoint decks, Parquet files - that land right on your desktop.\n\nAll of this runs in the browser using Pyodide. No local Python installation needed. The script covers patterns that most data workflows would need. Pick and adapt whichever patterns are relevant to your use case.\n\nResources\nExcel file with full script: Google Sheets\n\nxlwings Lite official site: lite.xlwings.org\n\nCricket data from Cricsheet, published under the Open Data Commons Attribution License 1.0 (ODC-BY).\n\n\nRelated\n\nTools: Practice Lab, Data Tools, DUCKIT - CSV to DuckDB\n\nExplore: Claude for Analytics hub, Database AI apps, Guides hub\n\nMore posts: DuckDB Meets Excel: xlwings Lite Data Tools, xlwings Lite Data Importer v2 Released, xlwings Lite new WINGMAN function - some usage patterns: python sandbox, stats, cleaning, bucketing, judging, How to Extract Python Code from xlwings Lite Excel Files"
},
{
"slug": "xlwings-lite-wingman-function-usage-patterns",
"title": "xlwings Lite new WINGMAN function - some usage patterns: python sandbox, stats, cleaning, bucketing, judging",
"tags": [
"xlwings-lite",
"python-in-excel"
],
"content": "xlwings Lite new WINGMAN function - some usage patterns: python sandbox, stats, cleaning, bucketing, judging\n\nWINGMAN\n\nFelix Zumstein released the new function just yesterday. Similar to =COPILOT in Excel. Tried out some common patterns. Wanted to get a feel for it.\n\nWorks wonderfully well.\n\nSharing my test file so you can see the usage: Google Sheets - WINGMAN test file\n\nQuick example patterns\n\nSome quick example patterns that I have tried out:\nHaving the LLM use its python sandbox\nPassing full range, adjacent ranges and single rows\nGetting back data in pre-specified structure so that it fits snugly next to my main table\nData cleaning - judgment based (non deterministic)\nCategorize customer profile based on data points - judgment based\n\nSetup\n\nSetup is straightforward - Settings - Local - Wingman. I tried it out with both OpenAI and Gemini. Works smooth. And option to bring custom model. So potentially you can use say an OpenRouter also...\n\nFelix's release post: LinkedIn\n\nxlwings Lite official site: lite.xlwings.org\n\n\nRelated\n\nTools: Practice Lab, AI Schema Detection, Technical Analysis in Excel\n\nExplore: Claude for Analytics, Browse all apps, Guides\n\nMore posts: xlwings Lite Local File Access: 8 Patterns You Can Use Today, Python In Excel - Claude Vs. xlwings Lite? Who Wins?, Python in Excel (xlwings Lite) with Natural Language Instructions."
},
{
"slug": "xlwings-utils-secure-cloud-access-vba-bridge",
"title": "xlwings_utils: Secure Cloud Access & VBA Bridge",
"tags": [
"xlwings-lite",
"python-in-excel"
],
"content": "xlwings_utils: Secure Cloud Access & VBA Bridge\n\nImport/export and manipulate images, PDFs, and binary files with xlwings Lite\n\nxlwings_utils by Ruud van der Ham\n\nRuud van der Ham has built xlwings_utils - solves two big problems I hadn't tackled. First: secure, private Dropbox access using OAuth tokens instead of shareable URLs. The same OAuth pattern works for Google Drive and GitHub (I tested both). No proxy, fully private for client data.\n\nSecond: local filesystem access via VBA bridge. xlwings Lite runs in browser sandbox - can't touch local files. His VBA bridge lets you import local images, manipulate PDFs with Python, export results back to local folder.\n\nSpent time digging through his code. Powerful solutions.\n\n▪️ The Dropbox approach: Secure & Private\n\nMy importer uses sharing links with CORS proxy. Works, but data passes through proxy. Ruud's approach: direct Dropbox API with OAuth refresh tokens. Set up tokens once as environment variables, that's it. I tested it works for Google Drive and GitHub also (GitHub needs proxy as it's not CORS enabled). More setup, but fully private.\n\nI'm adding this to my importer. Non sensitive data - shareable URLs. Private data - token-based API.\n\n▪️ The VBA bridge: This is the real breakthrough\n\nProblem: Can't import or export files, local images or PDF to local folder with xlwings Lite.\n\nRuud's solution: VBA macro bridge with base64 encoding. Python encodes files to base64, writes to Excel sheet chunks, VBA reads and decodes to local filesystem. Reverse works too.\n\nUse cases I'm testing:\nPDF Reports: ReportLab with matplotlib charts. Export directly to C:\\Reports\\. I was routing via cloud backend - this is way simpler.\nComplex Charts: 3D plots, polar charts, heatmaps. Generate in Python, export as PNG or insert into Excel/PowerPoint using openpyxl from within xlwings Lite.\nBatch Image Processing: Import 50 images via VBA bridge, process iteratively with PIL, export back.\nImport zip - unzip in xlwings Lite - process - zip - save back to local.\nTrigger email with attachment.\n\nBuilding baseline tools with VBA bridge for common workflows.\n\nOther utilities:\nstdout capture: Batch 1000+ print statements to memory, write once to Excel.\nVirtual filesystem helpers: Debug what files exist in Pyodide's virtual environment.\nopenpyxl integration: Convert between block structure and openpyxl worksheets.\n\nLinks:\nxlwingsutils: github.com/salabim/xlwingsutils\n\nSolid engineering. If you're doing serious work in xlwings Lite, check it out.\n\nBoth tools built on xlwings Lite - created by Felix Zumstein\n\nxlwings Lite official site: lite.xlwings.org\n\n\nRelated\n\nTools: Data Importer, Data Tools, DUCKIT - CSV to DuckDB\n\nExplore: Claude & Python in Excel hub, xlwings Lite Practice Lab, Build guides\n\nMore posts: xlwings Lite Local File Access: 8 Patterns You Can Use Today, xlwings Lite Data Importer v2 Released, DuckDB Meets Excel: xlwings Lite Data Tools, How to Extract Python Code from xlwings Lite Excel Files"
},
{
"slug": "yahoo-finance-mcp-backend-upgrade-jul2026",
"title": "Yahoo Finance MCP Backend Upgraded. 25 Tickers in One Request, Answers in Seconds, Clear Errors Instead of Slow Failures.",
"tags": [
"mcp"
],
"content": "Yahoo Finance MCP Backend Upgraded. 25 Tickers in One Request, Answers in Seconds, Clear Errors Instead of Slow Failures.\n\nGetting slow response on TigZig's Yahoo Finance MCP? Errors? Backend upgraded ... shiny new. You can now send up to 25 tickers in one request, it comes back in seconds instead of stalling, and if one name is bad or slow it returns a short note while the rest still comes through, with a clear message instead of a slow failure.\n\nIf you use the Yahoo Finance MCP on Claude, ChatGPT, LM Studio or anything else, it is vastly improved now.\n\nYou get an error on any TigZig apps ... it will pop up on my monitoring. I will hunt it down, find the root cause .. and if the issue is on my end, I'll fix it.\n\nUseful Links\nYahoo Finance MCP: https://www.tigzig.com/apis/yahoo-finance\nFor your AI agent, point it at https://api.tigzig.com\nIf anything errors on setup, the practical fix guide: https://www.tigzig.com/post/tigzig-api-errors-practical-guide-jul2026\nOn an older SSE URL, or new to connecting these: https://www.tigzig.com/post/tigzig-mcp-streamable-http-migration-jul2026\n\n\nRelated\n\nTools: Quants Agent, QRep - Security Reports, TREMOR - Macro Stress Signals\n\nExplore: API + MCP hub, MCP servers for AI agents, Quant apps hub\n\nMore posts: Every Tigzig MCP is now Streamable HTTP. Requesting existing users from LM Studio, ChatGPT, Claude etc to make a one time change if you have not already done so, Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users, Tigzig APIs - Reading the Logs, Shipping the Fixes. Status Page Live, R2 Downloads, Self-Correcting 4xxs., Pull your quants numbers directly with Claude / ChatGPT in Excel (with MCP)."
},
{
"slug": "yfin-api-csv-format-multiple-tickers-sep2026",
"title": "YFIN API Now Returns Plain CSV on Most Endpoints. Add format=csv to the Same URL.",
"tags": [
"portfolio-quants",
"excel"
],
"content": "YFIN API Now Returns Plain CSV on Most Endpoints. Add format=csv to the Same URL.\n\nYFIN API, my free Yahoo Finance API, now returns a plain CSV on most endpoints. Add format=csv to the same URL and get a normal table back, no JSON to unwrap.\n\nMultiple tickers. Every endpoint takes a comma-separated list of symbols, up to 25 in a single call, so a whole watchlist comes back in one request. Most used endpoints are prices, profile, financials and analyst price targets.\n\nDocs: tigzig.com/apis/yahoo-finance\n\nIf you would rather talk to an agent and have it pull the data for you, the data pull agent is sitting here: quants.tigzig.com\n\nAnd no, the data pull agent is not allowed to analyze the data and give a recommendation. It can just pull the data and generate it for you. However, it also generates an AI technical analysis report, with the technical analysis piece done by a different AI pathway. And yes, it can pull mutual fund NAVs also, for India only.\n\nYFIN, the free Yahoo Finance market data API: CSV on most endpoints, plus the full endpoint list"
},
{
"slug": "yfin-excel-webservice-429s-aug2026",
"title": "Getting 429s on the YFIN API from an Excel WEBSERVICE Formula? Here Is What Is Happening.",
"tags": [
"infrastructure",
"excel"
],
"content": "Getting 429s on the YFIN API from an Excel WEBSERVICE Formula? Here Is What Is Happening.\n\nGetting errors on TigZig's YFIN API with Excel WEBSERVICE formula while updating your trackers? Here is what is happening. WEBSERVICE is volatile, so it recalculates on any change anywhere in the sheet, and a few hundred of those cells will all fire at the same moment. My API has per-IP limits, so many of those come back as 429s, which means too many requests arriving at once.\n\nTwo things help. The endpoint takes up to 25 tickers in a single call, so a few hundred names becomes a few dozen calls. However with 100s of symbols that would still take a few minutes or more, as at the backend I pace the calls to manage concurrency and be polite to Yahoo Finance API.\n\nFor anything at that size I would use Power Query with M, or Python in Excel (xlwings Lite), where the calls run in order and refresh when you ask them to. My tool of choice for all such automations is xlwings Lite.\n\nYour AI, including the regular ChatGPT can help you with both options and it's fairly straightforward with AI. You paste the codes where it guides you and you are set. And if you want to go the Python in Excel (xlwings Lite) way you can check out my xlwings Lite hub tigzig.com/xlwings-starter for free guides and templates.\n\nI monitor the error rows. I have raised the limits at my end, and I am putting together simple templates for folks using YFIN APIs for updating trackers.\n\ntigzig.com\n\n\nRelated\n\nTools: DUCKIT - CSV to DuckDB, QRep - Security Reports, Quants Agent\n\nExplore: xlwings Lite hub, API + MCP hub, Claude for Analytics hub\n\nMore posts: TigZig API Rate Limits Raised Substantially, Across the Edge, the Proxy and the Apps. And a 429 Now Tells You What to Do About It., Zero Callers Rate Limited in 24 Hours. 6,500+ Calls and 9.7 GB Served, After the TigZig API Limits Went Up., Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users, Your Agents and Scripts Have Been Talking to Me. Here Is That Conversation in Plain English."
},
{
"slug": "yfin-integrate-your-tools-aug2026",
"title": "Pulling Yahoo Finance Data Into Your Own Trackers. Excel, Sheets, a Web Tool, or Straight From an AI Agent.",
"tags": [
"excel",
"portfolio-quants"
],
"content": "Pulling Yahoo Finance Data Into Your Own Trackers. Excel, Sheets, a Web Tool, or Straight From an AI Agent.\n\nNeed to pull Yahoo Finance market data into your trackers and tools? Excel, Sheets, Web tools? Prices, Calendar, Financials, Reccos. Tigzig's YFIN API is built for that. Share the API doc with your AI and have it set it up for you or guide you. Free, no signup, no key needed.\n\nIf your trackers are in Excel, you can pull the data via Power Query, Python-in-Excel (xlwings Lite), VBA or a WEBSERVICE formula. Each has trade-offs. There is a ready xlwings Lite tracker template for calendar and prices, and AI can adapt the code for you (link below).\n\nGoogle Sheets takes a few lines of Apps Script.\n\nIf you are building an HTML web tool, it's a simple API call from JavaScript.\n\nAnd if you work through an AI agent, it is on MCP and needs no code at all.\n\nYou don't need to code it yourself. If you are using an AI Agent / Coder it will do it for you, and if you are using AI Chat like ChatGPT it can share code / formula and guide you where to paste. Share or copy the API docs with it, explain what you need.. it knows what to do.\nAPI docs (for you): tigzig.com/apis/yahoo-finance\nFor your AI: yfin-h.tigzig.com/v1/openapi.json\nPython-in-Excel (xlwings Lite) tracker template: An Excel Tracker for Hundreds of Yahoo Finance Tickers\n\nEasiest is with an AI agent\n\nEasiest is with an AI agent, Claude Cowork, ChatGPT, or a coder like Claude Code or Google Antigravity. Basic subscriptions cover these.\n\nYou need the desktop app, it does not work on web. The agent goes straight into your Excel file. An xlsx is XML underneath, so it works on the internals directly. It will set up the Power Query step, write the formula, or lay out the sheet the way you asked. You don't need to know the code.\n\nPython-in-Excel (xlwings Lite) is what I mostly use. I describe what I want, it gives me the code, I paste it in the code box and run it.\n\nWEBSERVICE it will also set up for you. That one is volatile though, all the cells fire at once, so ask it to batch, yfin takes 25 tickers in a call. For anything large or frequent, Power Query or xlwings Lite is the better route.\n\nSame idea for Google Sheets, or a JavaScript web tool.\n\nNo agent subscription? Plain chat is fine. Share the API doc link, ask it to use the batch API, and ask it to show you where to paste. A bit more tedious but works.\n\nEither way, validate what comes back.\n\nQuick start guides: tigzig.com/claude-for-analytics\n\nA few quick ones\nYahoo wants the exchange ticker, not the company name. For Indian stocks that is RELIANCE.NS on the NSE, RELIANCE.BO on the BSE. The company name with .NS on the end returns nothing. You can check the symbol on the Yahoo Finance site or use the search endpoint (see below).\nThere is a search endpoint now, and it is mainly there for agents. Point your agent at it and it stops guessing tickers from memory. It searches, gets back the real symbols with the exchange each one sits on, and asks you which one you meant. You can also set up a search box in your tool off it. One caution: Yahoo always returns its nearest guesses, so a name that does not exist still comes back with confident looking rows, sometimes an index or a futures contract. Check the match.\nFor the financial statements, use the /v1/excel/ versions. They come back as structured rows that go into a sheet cleanly. The plain ones are pipe delimited text.\nOn WEBSERVICE, ask your agent for an array formula that sends 25 tickers per call instead of one cell per ticker. A few hundred single-ticker cells all recalculate at the same instant, arrive as one burst, and start timing out.\n\ntigzig.com/apis/yahoo-finance\n\n\nRelated\n\nTools: DUCKIT - CSV to DuckDB, QRep - Security Reports, TREMOR - Macro Stress Signals\n\nExplore: Yahoo Finance API docs, Python in Excel with xlwings Lite, Claude for Analytics hub\n\nMore posts: An Excel Tracker for Hundreds of Yahoo Finance Tickers, Built with xlwings Lite. For When WEBSERVICE Keeps Failing., Three Fixes on the Yahoo Finance API This Week, All of Them Found by Reading My Own Error Logs., Getting 429s on the YFIN API from an Excel WEBSERVICE Formula? Here Is What Is Happening., Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users"
},
{
"slug": "yfin-three-fixes-aug2026",
"title": "Three Fixes on the Yahoo Finance API This Week, All of Them Found by Reading My Own Error Logs.",
"tags": [
"portfolio-quants",
"excel"
],
"content": "Three Fixes on the Yahoo Finance API This Week, All of Them Found by Reading My Own Error Logs.\n\nTigzig's Yahoo Finance API is the 2nd largest in the platform, averaging close to 2500 successful API calls per day. Three fixes went into this week, and all three came out of reading my own error logs.\n\nThe one that bothered me most: price history was failing and then telling the user their ticker symbol or their date range was wrong, when both were usually fine. The real cause was upstream on Yahoo Finance side. Someone at the other end would have spent a while re-typing a symbol that was correct all along. Now fixed.\n\nThe other two were similar.\n\nDetailed company info could run a long time and then give you nothing back, losing the tickers that had already come in fine... it now returns whatever was fetched.\n\nAnd if somebody put even a single ISIN in a symbol list ..that used to stop the whole request...now that is skipped and the rest of your list still comes back.\n\nYour API calls talk to me. You hit an error, I see it. Your calls go slow, I see it.\n\nIf it's on my side, I will fix it, some quickly some might take a little while more.\n\nIf you were hitting errors last week, try again. Mostly likely they are gone.\n\nFree, no signup, no key.\n\ntigzig.com ➜ API/MCP ➜ Yahoo Finance\n\nFor users preparing Excel trackers with WEBSERVICE\n\nFor users preparing Excel trackers using WEBSERVICE, it is a volatile formula, so it recalculates on any change anywhere in the workbook and a few hundred of those cells all fire at once, which is what produces the rate-limit you might sometimes see.\n\nThere is a downloadable Excel workbook that pulls dividend and earnings calendar data and prices for hundreds of tickers through this API, built with xlwings Lite, which runs Python inside Excel. It batches at up to 25 tickers a call and writes each batch back to the sheet as it arrives, so you can watch it progress. It ships set up for the S&P 500 and the ticker list is swappable.\n\nAn Excel Tracker for Hundreds of Yahoo Finance Tickers\n\nYou don't need to know python to use it. You can just handover the embedded Python code to your AI and it will adapt it for you. For more guides on xlwings Lite go to xlwings-lite.tigzig.com.\n\nAnd there is a guide covering what each error from these APIs means and what to do about it, across the mutual fund NAV, macro and Yahoo endpoints: Demystifying TIGZIG API Errors\n\nA common error with Indian stocks\n\nOne common error I see is with Indian stocks - Yahoo wants the exchange ticker, not the company name.\n\nSo RELIANCE.NS works and RELIANCEINDUSTRIES.NS returns nothing. Same for the BSE side with .BO.\n\nQuickest way to check one you are unsure about: search the name on Yahoo Finance itself and copy the symbol exactly as it appears there, including the case. If a symbol works on the Yahoo website it will work here, because this is the same source.\n\n\nRelated\n\nTools: DUCKIT - CSV to DuckDB, QRep - Security Reports, TREMOR - Macro Stress Signals\n\nExplore: Yahoo Finance API docs, Python in Excel with xlwings Lite, API and MCP catalog\n\nMore posts: An Excel Tracker for Hundreds of Yahoo Finance Tickers, Built with xlwings Lite. For When WEBSERVICE Keeps Failing., Demystifying TIGZIG API Errors: A Practical Guide for MFPRO, TREMOR, and VIGIL Users, Getting 429s on the YFIN API from an Excel WEBSERVICE Formula? Here Is What Is Happening., Yahoo Finance MCP Backend Upgraded. 25 Tickers in One Request, Answers in Seconds, Clear Errors Instead of Slow Failures."
},
{
"slug": "your-api-key-is-visible-in-the-browser",
"title": "Your API Key Is Visible in the Browser. Even if you put it as Vercel's 'secret' backend env variable.",
"tags": [
"security",
"infrastructure",
"fastapi"
],
"content": "Your API Key Is Visible in the Browser. Even if you put it as Vercel's 'secret' backend env variable.\n\nThe Power of Vercel Serverless Functions\n\nQuick version: There are a few common things people try to hide API keys in frontend apps that don't actually work. And one thing that does. This post walks through each scenario so you know exactly where you stand.\n\nThe Problem\n\nYour app needs to call an external API - OpenAI, a database, your own backend, whatever. That API needs a key. And if anything about that call happens in the browser, the key is findable. Maybe not obvious, maybe needs two clicks in DevTools, but findable.\n\nWhat Doesn't Work\n\nAttempt 1 - Put the key directly in your HTML or React component code\n\nThis one is obvious once you think about it. The key is right there in your source or in the network tab when the call fires. Anyone who opens DevTools -> Network tab can see the Authorization header with your key in plain text.\n\nAttempt 2 - Use a React / Vite environment variable with the public prefix (REACTAPP or VITE_)\n\nThis is the most common mistake. People read \"environment variable\" and think \"hidden.\" But any variable with that prefix gets bundled directly into the JavaScript that ships to the browser at page load. It is sitting in a JS file anyone can download and read with a text editor. Not hidden at all.\n\nAttempt 3 - Use a secret environment variable without the prefix, but still make the API call from the frontend\n\nThis one is more subtle. Without the public prefix, Vercel does keep the variable server-side - it never gets bundled into the JS. So the variable itself is not exposed. But you are still making the API call from your frontend code. And when that call fires, the Authorization header with your key shows up in the network tab. The variable was hidden but the key got out anyway through the actual HTTP request.\n\nSo the rule is simple: if the API call happens in the browser, the key is visible. Doesn't matter how you stored the variable.\n\nWhat Works - The Serverless Function\n\nThe fix is to move both things to the server side - the environment variable AND the API call.\n\nA serverless function is basically a mini backend that runs on the server. Think of it like a mini FastAPI or Flask - but running JavaScript, and you don't have to manage any server yourself. Vercel gives you this for free.\n\nAnd the beauty of it - there is nothing to install or configure. You create a folder called /api in your project root, drop a JS file in it, and Vercel automatically runs it on the server. That folder is your mini backend. Whatever goes in there never runs in the browser. It's like JavaScript version of a FastAPI server - attached to your app.\n\nYour frontend calls /api/your-function. That function reads the secret variable and makes the real API call. The browser's network tab only shows the call to /api/your-function - the downstream API call with your key is completely invisible because it happens on the server.\n\nBasic pattern:\n\n``javascript\n// WRONG - key visible in network tab\nconst response = await fetch(\"https://api.openai.com/v1/chat\", {\n headers: { \"Authorization\": \"Bearer sk-your-key-here\" }\n});\n\n// RIGHT - frontend calls your own serverless function\nconst response = await fetch(\"/api/ask\", {\n method: \"POST\",\n body: JSON.stringify({ question: userInput })\n});\n`\n\n`javascript\n// /api/ask.js (runs on Vercel server, never in browser)\nexport default async function handler(req, res) {\n const response = await fetch(\"https://api.openai.com/v1/chat\", {\n headers: {\n \"Authorization\": Bearer ${process.env.OPENAI_KEY} // server-side only\n },\n body: JSON.stringify(req.body)\n });\n const data = await response.json();\n res.json(data);\n}\n``\n\nThe OPENAI_KEY is set in Vercel's dashboard environment variable settings - no prefix needed, stays server-side. The browser never sees it, the network tab never shows it.\n\nCommon Scenarios\n\nPlain HTML app on Vercel\n\nEven a single HTML file can use this pattern. Just add an /api folder alongside your HTML. The HTML calls /api/your-function. Vercel handles the rest. You don't need React or any framework.\n\nReact app on Vercel\n\nSame thing. Your React components call /api/your-function. The function handles the API call with the secret key. Variable in Vercel dashboard, no prefix.\n\nFlask or FastAPI backend\n\nHere the backend itself is your \"serverless function\" equivalent. Your HTML or React frontend calls your Flask/FastAPI route. That route reads the key from the server's environment variable and makes the API call. Key never touches the browser. The difference from Vercel serverless is you are managing your own server - Hetzner, EC2, whatever. More control, more work. For simple use cases, Vercel serverless is easier. For heavier workloads or existing backends, Flask/FastAPI makes more sense.\n\nThe Power of Serverless Functions - What Else Can You Do?\n\nRate Limiting\n\nSince the serverless function is the first thing that sees the request, you can rate limit right there before anything else happens. Upstash Redis is the standard option - it is a key-value store, free tier is enough for most apps. You store a counter per IP address, increment it on every request, block if it crosses your limit. And the IP you get at the Vercel serverless layer is the real TCP IP - not X-Forwarded-For which can be faked with a curl command. Vercel sees the actual connection. So even if someone runs an automated attack and tries to spoof their IP via headers, you are rate limiting the real IP. We covered this in detail in the SlowAPI post linked below - same concept applies here.\n\nData Processing\n\nThe serverless function is not just a pass-through. You can do real work in there. Call an API, get the data back, clean it, reshape it, filter it, before sending the result to your frontend. On Vercel's free plan you now get 300 seconds per function invocation - that is 5 minutes, which is a huge amount for most API calls and light processing tasks. So it is genuinely a mini backend, not just a proxy. Fair amount of work can get done there without spinning up a separate server.\n\nSimple Security Gate\n\nYou can add a basic security check in the serverless function - check for a hardcoded token or a simple passcode before doing anything. Not the strongest protection, it is subject to brute force if someone is determined. But it stops casual abuse and random probing immediately. You can harden it further by locking after a few failed attempts. For anything serious you still want proper OAuth or Google auth - but as a first gate, a simple token check in the serverless function is easy to add and better than nothing.\n\nWhat's the Catch?\n\nIt times out at 300 seconds - that's 5 minutes and still quite a lot. For operations that exceed this - large file uploads, heavy database imports, long-running data processing - use one-time tokens, signed URLs, JOB ID + Polling. That's a different topic in itself.\n\nOne More Thing - This Is Not Complete Security\n\nHiding the key is the bare minimum. But your /api/your-function URL is still visible in the browser. Anybody can call it directly, bypassing your frontend. So now the serverless function is your attack surface.\n\nSecurity is always multi-layered - at the edge with Cloudflare, OAuth, API Keys, Rate Limits in serverless functions, database security, backend security etc. Each layer adds protection. None of them alone is the full answer. The serverless proxy is just one layer.\n\nPractical Note for People Working With AI Coders\n\nAI coders - if you don't specify, they will keep the API Key in environment variable but put the API call in the frontend. That's of no use. Tell your AI coder upfront: all API calls go through a serverless function, no keys in the frontend. And ask it to add basic rate limiting while it is at it. Much easier to do at build time than to retrofit later.\n\nRelated\nAre You Rate Limiting the Wrong IPs? A SlowAPI Story - How multi-hop architectures cause rate limiters to throttle the wrong IP addresses\nFull security checklist for web apps - 95 items across React, FastAPI, Postgres, DuckDB, Cloudflare, MCP servers, Auth and VPS security. Plain English, with code fixes. Download as markdown and paste to your AI coder.\n\n\nRelated\n\nTools: TREMOR - Macro Stress Signals, Quants Agent\n\nExplore: Security checklist for web apps, APIs and MCP hub, Guides hub\n\nMore posts: Are You Rate Limiting the Wrong IPs? A SlowAPI Story., Upstash Redis on Vercel - The Tool I Didn't Know I Needed, 2026 Infra Guide for AI Tool Builders - Part 3: The 18 Common Security Mistakes and How to Fix Them, Going beyond Google Login for critical apps. Identifying gaps & hardening your entry points., 2026 Infra Guide for AI Tool Builders - Part 5: Perimeter Security on Cloudflare (Free Plan)"
}
]
}