# VIGIL Data Validation - Reconciliation & Audit Approach The two-level validation behind VIGIL: automated integrity checks on every data update, plus periodic independent audits reconciling the database against the original NSE sources - with the headline results and findings for each dataset. Source data: VIGIL reconciliation audits vs NSE source files and web pages. Last updated: 2026-07-02. Interactive tool: https://vigil.tigzig.com/validation VIGIL validates every dataset on two levels. First, **automated integrity checks run on every data update** - threshold checks that reject a truncated or corrupt source file, atomic replacement (the database is only updated if all checks pass, otherwise existing data is preserved), deduplication, and full sync logging (timestamp, status, record counts) for an audit trail. Second, **periodic independent audits** reconcile the database against the original NSE source files and live NSE web pages, with every finding published. This page summarizes the validation approach and the headline audit results per dataset; each links to its detailed methodology page. [Open the live VIGIL Validation view](https://vigil.tigzig.com/validation) for the full per-dataset reconciliation tables. ## Surveillance (ASM / GSM / ESM) Audit of 9 March 2026: the REG_IND source CSV (2,807 total securities, 398 flagged) was downloaded from NSE archives and parsed independently; each of the 11 flag categories was counted in the raw CSV and compared to the database - **100% match, all 11 categories, 398 = 398**. A sample was cross-verified against the live ASM and GSM pages on nseindia.com (Long Term ASM, Short Term ASM and GSM symbols all matched). Two findings: two GSM symbols (ASIL, ATNINTER) appeared on the NSE web page but not the 6-March CSV (a normal one-trading-day lag), and 6 of the 17 REG_IND flag columns are intentionally not captured (4 of them currently have zero flagged companies). Update-time checks require more than 1,000 securities and more than 100 flagged before any update is permitted. See [Surveillance methodology](/vigil/surveillance). ## Credit Ratings Audit of 9 March 2026: 10,475 records, **zero duplicate record IDs**. The NSE Centralised Rating Database API returns heavy duplicates (28-75% per batch); VIGIL's 17-field MD5 deduplication rejects them so each unique field combination gets one record. An independent cross-check pulled raw API data for 3 recent dates and compared record-for-record: **3 of 3 dates matched** after dedup. Field completeness is ~100% (credit_rating 99.8% - the 18 nulls are legitimate "Other" actions; outlook 96.6% - some actions carry no outlook). Every record is classified against a configurable red-flag map, and debt ISINs are enriched to their parent equity ISIN via a three-pass lookup. See [Data Sources](/vigil/data-sources). ## Related Party Transactions A full extraction validation in March 2026 across all 1,440 XBRL filings (728 companies, ~250,000 transactions) independently counted the RPT contexts in each XBRL file and compared them to the parser's output: **100% match across all 1,440 filings, zero discrepancies**, with all required fields populated and no duplicate transaction numbers within a filing. See [RPT methodology](/vigil/rpt). ## Quarterly Results An independent 4-phase validation on 19 March 2026 covering 458,000+ data points across 15 quarters: source-file verification (72 company-quarters, 4,176 values re-parsed from the original XBRL - 100% match), 19 database-integrity checks (15 passed outright, 4 issues remediated), filing-revision verification (all 751 companies confirmed on the latest version), and frontend-calculation verification (25 calculations re-computed - 100% match). Documented unit-mismatch and insurance-format exclusions are published in full. See [Quarterly Results methodology & validation](/vigil/quarterly-results). ## Insider Trading, SAST, Pledge and Encumbrance These NSE disclosure feeds follow the same validation pattern: source-file threshold checks and atomic replacement on every update, deduplication against prior records, and sync logging, plus periodic reconciliation of the database against the NSE source. Each dataset's definitions, coverage and filing rules are documented on its own page: [Insider Trading](/vigil/insider-trading), [Takeover (SAST)](/vigil/sast), [Promoter Pledge](/vigil/pledge), and [Encumbrance events](/vigil/encumbrance). ## BRSR (ESG) BRSR is validated differently because the values are self-reported ESG disclosures. VIGIL runs a layered consistency-check system (algorithmic cell-level checks across 52 fields) plus a printed-PDF validation layer that catches cells which pass every algorithmic check but disagree with the signed PDF. The honest bottom line: aggregate and directional analysis is highly reliable, but a specific company cell used for a specific decision should be verified against the PDF. Full detail on the [BRSR data quality](/vigil/brsr-data-quality) page. [Open the live VIGIL Validation view on TIGZIG](https://vigil.tigzig.com/validation) for the full reconciliation tables, or see [all VIGIL data sources](/vigil/data-sources). --- Source: https://www.tigzig.com/vigil/data-validation