parser broke on the quarterly business review PDF: agent extracted wrong renewal dates and the sync failed

TL;DR

QBR decks are presentation files, not data files: dates appear as quarter labels, relative phrases, and formatted tables that a naive extractor reads in layout order, not logical order. Extract with layout awareness, parse every date pattern present in real decks, and validate each extracted renewal date against the CRM; mismatches go to a review queue instead of into the sync.

parser broke on the quarterly business review PDF: agent extracted wrong renewal dates and the sync failed

Use this when

• Parsed renewal dates are wrong for a subset of accounts
• Errors cluster on certain deck templates or presenters
• The sync wrote the wrong dates into downstream systems

Not for this skill when

• The decks themselves contain wrong dates (a human authoring error)
• The CRM dates are the wrong ones and the decks are right

Steps

1. Collect the decks that produced wrong dates and find the exact text and position of the true date in each. Verify: the extraction failure pattern (pattern, order, or table split) is characterised
2. Upgrade extraction to preserve layout (column and row position) for table-based dates. Verify: dates extract with their account context attached
3. Parse all observed date patterns, including quarter-end phrases resolved against the deck's date. Verify: the previously failing decks now yield correct dates
4. Validate every extracted date against the CRM and contract system, queuing mismatches for human review. Verify: no unverified date reaches the sync
5. Repair the dates already synced wrongly and notify the teams that consumed them. Verify: downstream forecasts and plans are corrected at the source

Variant phrasings

QBR PDF wrong renewal dates extracted

A related phrasing for the same underlying problem.

PDF date parsing layout errors

A related phrasing for the same underlying problem.

validate parsed dates against CRM

A related phrasing for the same underlying problem.

Compatibility: Compatibility: PDF extraction tooling (pdfplumber, PyMuPDF, or OCR for scanned decks) feeding CRM syncs, current versions.

Why it happens

Presentation PDFs encode meaning in position, so text-order extraction scrambles exactly the associations that matter (which date belongs to which account). Layout-aware extraction plus external validation converts an untrustworthy narrator into checked data.

Edge cases / pitfalls

• Scanned or image-based decks need OCR, and OCR confuses similar digits in dates. Validation against the CRM is doubly important there.
• Quarter phrases like Q3 renewal are ambiguous between calendar and fiscal years. Resolve using the customer's fiscal calendar from the CRM.
• A review queue needs a service-level expectation, or mismatches pile up unsynced and someone reverts to trusting the parser.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_XGxqT3XlgKbSI7H2iFtADQ