Three causes, and only one of them is your parsing. Sum every holding-level market value in a monthly disclosure, compare it against the grand total the fund house prints in the same file, and the two disagree often enough that it is worth knowing which of these you are looking at before you go hunting for a bug.
1. Derivatives, included by you and excluded by them. Some houses leave the mark-to-market on stock and index futures out of their own printed grand total, while a straight sum of every line includes it. Your figure then comes out higher, by a fraction of a per cent. In one published reconciliation the gap ran between 0.19 and 0.88 per cent across eleven fund-months, all of them one house, and the largest single month traced to one stock future.
2. Where the house puts cash. This is the one that changes the denominator of every percentage on your screen. Some houses put cash and TREPS inside the holdings list, so the total at the bottom is already the whole fund. Others print a securities total first and cash on a line below it, and that total is short by whatever the fund is holding in cash. Take the net assets figure in both cases, and a percentage is then a share of the whole fund whichever house filed it.
3. The printed total is sometimes typed, not calculated. A grand total cell can be a hardcoded value rather than a sum formula, and then it is simply wrong. Two months of one large fund were short by about Rs 15 lakh and Rs 84 lakh against a portfolio of roughly Rs 1.2 lakh crore - a rounding error in proportion, and an arithmetic error in their workbook rather than in the parse.
How to tell your parse is fine. Check your sum against the disclosure's own section sub-totals, not only against its grand total. If every section reconciles and only the grand total does not, the difference is a methodology choice or a source-file error, and the right response is to publish it with the amount and the reason rather than quietly adjust a number a regulator-filed document states.
The full reconciliation is published, pass and fail: 251 of 264 fund-months match to tolerance across eleven monthly snapshots, and the thirteen that do not are each listed with the amount and the cause - https://www.tigzig.com/mfpro/composition-validation. The tool: https://www.tigzig.com/mfpro (Composition). Movement that looks like a trade and is not: https://www.tigzig.com/agents-faq/why-does-a-fund-holding-change-when-the-fund-did-not-trade.
Building something like this? How I work covers the rates, the availability and what I take on.