the agent's diff flagged the entire signature block as deleted - counterparty switched Letter to A4 and pagination...

TL;DR The counterparty switched from Letter to A4, every page break moved, and a page-anchored diff flagged the entire signature block as deleted. Diff at the paragraph or token level with page breaks stripped out, and handle signature blocks as structured content rather than flowing text. Repagination is not a deletion.

What the bogus diff reported:

- [signature block: 2 signatories, 4 signature lines] marked DELETED

Steps

1. Confirm the deletion is repagination. Find the signature block text in the new version.

Command: grep for a signatory name from the old signature block in the new document text. Expected: the signatory names exist in the new version, just on a different page, proving nothing was deleted.

1. Strip page breaks and normalize pagination artifacts before diffing: remove form feeds, page-number headers and footers, and collapse the text into a continuous paragraph stream.

Expected: re-running the diff shows the signature block as unchanged.

1. Extract signature blocks as structured records, signatory name, title, date lines, rather than diffing them as prose. Compare the records field by field.

Expected: the signature block comparison reports field-level status, like both signatories present, instead of a wall of deleted lines.

1. Add a page-size check to your version-intake step: record page dimensions of each version and warn when they differ, so repagination is expected before the diff runs.

Expected: future Letter to A4 switches produce a "pagination changed" note, not a deletion report.

Use this when

• The counterparty changed page size, margins, or orientation between versions
• Whole blocks like signatures, exhibits, or schedules show as deleted but exist in the new version
• Page breaks shifted and your diff anchors content to page numbers
• A "no real changes" version still produces a long redline

Not for this skill when

• The signature block was actually removed or a signatory changed, that is a real change worth flagging
• Page breaks are meaningful, like a contract that requires each section to start on a new page
• The diff is page-image based, where layout is the content, use visual comparison instead
• Only a few lines shifted, check them by hand rather than retooling

Variant phrasings

• diff flagged signature block as deleted after Letter to A4 switch
• pagination shift made the redline report phantom deletions
• counterparty changed page size and the diff shows everything moved
• how to diff contracts ignoring page breaks
• signature block shows as deleted but it is just repagination

Why it happens

Many diff pipelines keep page breaks as hard boundaries and compare page by page, so when A4's extra length shifts every break, content lands on different pages and the page-anchored comparison sees deletions on the old pages and additions on the new ones. The signature block is the most visible victim because it sits at the end where the cumulative shift is largest, but the same phantom deletion hits every page boundary.

Edge cases

• A real deletion inside a repaginated document is easy to miss once you stop trusting page-anchored diffs, so always review the paragraph-level change list, not just the move report.
• Scanned versions may also change DPI, which shifts OCR line breaks on top of the page-size shift, normalize both.
• Signature blocks sometimes gain a signatory between versions, field-level comparison catches this where prose diffing buries it.
• If the counterparty also reordered sections, combine this fix with move detection so reordered content is not misread as deleted.
• Keep the page numbers from each version in your citations even though the diff ignores them, reviewers still navigate by page.

Provenance

Resolved from the public thread: https://vectle.com/posts/pstLCuov8UXypU3tHGwWpGvg