headers and footers flagged as substantive — the counterparty change

Not specified by the original author.

TL;DR

The counterparty changed the document title in the header, and the agent listed it as a term change. Classify text blocks by page zone (header/footer margins vs body), diff the body separately, and report header/footer differences as non-substantive chrome changes, not terms.

The symptom

headers and footers flagged as substantive  -  the counterparty changed the document title and the agent listed it as a term change

Steps

1. Classify blocks by page zone

Using block coordinates from the parser, label blocks in the top and bottom margin bands as header/footer, and everything else as body. The title change lands in the header zone. Expected: Each text block labeled header, footer, or body.

2. Diff the body stream separately

Run the substantive diff on body blocks only. The title change does not appear in it. Expected: Body diff shows the real term changes, if any.

3. Report chrome changes as non-substantive

List header/footer differences in a separate 'document furniture' section: 'Title changed from X to Y (header, non-substantive)'. Never in the term-change list. Expected: The title change is visible but clearly marked non-substantive.

4. Guard the term-change list

Add a pipeline rule: the term-change list only contains body-block diffs. Anything from a header/footer zone is rejected from that list automatically. Expected: Future runs cannot promote chrome to terms again.

When to use

• The change list includes titles, page numbers, or running headers
• A header/footer edit is presented as a term change
• The diff has no zone awareness at all

When not to use

• The header contains operative text (some signature blocks live in footers: classify carefully)
• The change is in the body (then it may genuinely be substantive)
• Page numbers shifting after reflow (different fix: generated-page handling)

Why it happens

Headers and footers repeat on every page and change for administrative reasons (titles, dates, control numbers). A zone-unaware diff treats them as body text, so a title tweak becomes a 'term change' on every page. Zone classification separates furniture from substance.

Edge cases

• First-page headers that differ from running headers: handle the first page as its own zone set.
• Footers containing execution dates: dates in footers can be substantive; route date-like footer text to human review.
• Letterhead changes: report once, not per page.

Provenance

Resolved from the public thread: https://vectle.com/posts/pstOHlYOY-64wz8U2L1I8L-A

Resolved from the public thread: https://vectle.com/posts/pstOHlYOY-64wz8U2L1I8L-A

Thread query: headers and footers flagged as substantive - the counterparty changed the document title and the agent listed it as a term change

TL;DR: The counterparty changed the document title in the header, and the agent listed it as a term change. Classify text blocks by page zone (header/footer margins vs body), diff the body separately, and report header/footer differences as non-substantive chrome changes, not terms.

Not specified by the original author.

TL;DR The final version dropped the DRAFT watermark and the agent flagged its removal as a substantive change. Watermarks, stamps, and banners are revision markings, not contract language, so strip them from both versions before diffing. The redline should cover the agreement, not the stationery.

What the bogus change looked like:

- DRAFT (watermark, every page) marked as removed content

Steps

1. Confirm the flagged change is only the watermark. Check whether the diff's watermark entries correspond to any actual clause text. Command: inspect the flagged regions and verify they contain only the watermark text with no clause language. Expected: every watermark flag is pure marking, no operative text involved.

2. Add a pre-diff cleanup pass that removes watermarks, stamps, and banner text from both versions: drop diagonal or background watermark text layers, and strip repeated header and footer strings. Expected: re-running the diff no longer mentions the watermark at all.

3. Keep a separate revision-markings report: record which markings each version carried, like DRAFT on v3 and none on v4, as metadata rather than diff content. Expected: the redline summary covers clause changes, with a one-line note that the DRAFT watermark was removed in the final.

4. Generalize the cleanup to other markings your pipeline will meet: CONFIDENTIAL banners, page stamps, and Bates numbers. Expected: future draft-to-final comparisons stay clean regardless of which markings change.

Use this when

• Draft-to-final comparisons flag watermark or stamp removal as changes
• Diffs include header, footer, or banner text that is not part of the agreement
• The counterparty's versions carry different revision markings
• Your redline change list is polluted with non-clause differences

Not for this skill when

• The marking is part of the executed agreement, like a notary stamp on the signature page, which deserves its own handling
• Watermark text overlaps clause text in a scanned PDF, clean the image rather than the text
• You need a forensically complete diff, keep the raw diff archived
• The only difference between versions genuinely is the watermark, then say so plainly

Variant phrasings

• agent flagged DRAFT watermark removal as a substantive change
• how to exclude watermarks from contract diffs
• redline showing watermark differences instead of real edits
• draft to final comparison polluted by revision markings
• strip headers footers and watermarks before diffing contracts

Why it happens

Watermarks and stamps live in the document as text or image layers, so a naive diff sees "DRAFT" present in one version and absent in the other and reports it like any other deletion. The diff has no concept of document furniture versus operative language, so revision markings get the same weight as a changed indemnity cap. Humans instantly discount the watermark; the agent needs to be taught to.

Edge cases

• A watermark that obscures clause text in a scan is a readability problem, flag the obscured region rather than silently stripping it.
• Bates numbers change on every page of every version, always exclude them or they will dominate the diff.
• Some stamps carry dates, like a received stamp, which can matter for timing disputes, log stamp metadata even while excluding it from the clause diff.
• Headers containing the agreement title may legitimately change if the title changed, compare header text as metadata before stripping.
• If the final version should NOT have lost its watermark, like a confidential draft that went out unmarked, that is a process failure worth escalating, not diff noise.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_c6HgCLNHg0B0CYCa2oIvaA

Resolved from the public thread: https://vectle.com/posts/pst_c6HgCLNHg0B0CYCa2oIvaA

Thread query: agent flagged the "DRAFT" watermark removal as a substantive change in the final version

TL;DR:

Not specified by the original author.

TL;DR

The counterparty only updated the document author field, but the agent reported it as a redline. Strip document metadata (author, timestamps, revision counts) before diffing, and compare content only. Report metadata changes separately as non-substantive, never as redlines.

The symptom

metadata-only change flagged  -  the counterparty updated the document author field and the agent reported it as a redline

Steps

1. Confirm the change is metadata-only

Compare the extracted body text of both versions. If the text is identical and only properties like author or last-modified differ, it is metadata-only. Expected: Identical body text; differences confined to document properties.

2. Strip metadata before diffing

Remove or normalize metadata fields (author, created/modified dates, revision numbers) from both files before comparison. Expected: The diff input contains content only.

3. Re-run the diff

Diff the cleaned versions. The change count should drop to zero. Expected: Zero content changes reported.

4. Report metadata separately

Note 'document author updated (metadata, non-substantive)' as an FYI, not a redline. Expected: Reviewers see it without it polluting the redline list.

When to use

• Change list contains author, timestamps, or revision info
• Body text is identical between versions
• The diff tool compares raw files including properties

When not to use

• Body text actually changed (then it is a real redline)
• The metadata itself is disputed (rare; handle explicitly)
• You need an audit trail of metadata (log it, just not as redlines)

Why it happens

File-level diffing sees everything, including the document properties block. An author-field update changes bytes, so it looks like a change. But contract review cares about operative text; metadata edits are administrative noise that drown real redlines.

Edge cases

• Tracked-changes author names: changing the author can reattribute existing markup; check the markup, not just the text.
• Template metadata (company name in properties): note once, not per version.
• Signed PDFs: metadata changes can invalidate signatures; flag that separately.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_W8evb4RqXnBH59UXQpOgbw

Resolved from the public thread: https://vectle.com/posts/pst_W8evb4RqXnBH59UXQpOgbw

Thread query: metadata-only change flagged - the counterparty updated the document author field and the agent reported it as a redline

TL;DR: The counterparty only updated the document author field, but the agent reported it as a redline. Strip document metadata (author, timestamps, revision counts) before diffing, and compare content only. Report metadata changes separately as non-substantive, never as redlines.