agent's redline summary said "47 changes" but 46 were smart-quote replacements from the other side's Word auto-correct

TL;DR Word's auto-correct replaced every straight quote with a curly quote, and the diff counted each one as a change, turning 1 real edit into 47 reported changes. Normalize all quotation marks to one canonical form before diffing, and report cosmetic changes separately from substantive ones. Punctuation normalization is not a substantive edit.

What the inflated summary looked like:

47 changes found: 46 quote character replacements, 1 wording change

Steps

1. Confirm the changes are cosmetic. Sample the flagged changes and check whether only the quote characters differ.

Command: diff the two versions and inspect whether the changed lines differ only in quote glyphs. Expected: the vast majority of changes are straight quotes becoming curly quotes with identical surrounding words.

1. Add a quote-normalization pass before diffing: map curly double quotes, curly single quotes, and prime marks to their straight ASCII equivalents on both versions.

Expected: re-running the diff drops the change count from 47 to the real number, in this case 1.

1. Split your redline report into substantive and cosmetic sections. Run the diff twice, once on normalized text for the substantive list, once on raw text, and report anything that only appears in the raw diff as cosmetic.

Expected: the summary reads "1 substantive change, 46 cosmetic (quote normalization)" instead of "47 changes".

1. Add the other side's auto-correct artifacts to the same normalization: em dashes from double hyphens, ellipsis characters, and non-breaking spaces.

Expected: future Word-roundtripped versions produce clean diffs with no cosmetic inflation.

Use this when

• A redline change count looks far too high for the actual revision
• Diffed lines differ only in quote characters, dashes, or whitespace glyphs
• The counterparty edits in Word with auto-correct and auto-format enabled
• Your redline summary gives one number with no cosmetic versus substantive split

Not for this skill when

• The quote change is legally meaningful, like a defined term gaining quotes for the first time, which deserves a flag
• The changes are real wording edits that happen to sit near quotes
• You need a character-exact redline for litigation, keep the raw diff as the record
• The document never passes through Word, check your own pipeline for the normalization gap

Variant phrasings

• redline says 47 changes but 46 are smart quotes
• Word auto-correct curly quotes inflating the diff count
• how to ignore smart quote replacements in contract diffs
• cosmetic changes flooding the redline summary
• diff counts quote character swaps as substantive changes

Why it happens

Word silently converts straight quotes to curly quotes as you type, so a version that passed through Word carries different quote glyphs on every quoted passage even when nobody edited a word. A byte-level diff sees each glyph swap as a change, and the summary counts them all equally. The diff is not wrong about the bytes, it is just answering the wrong question: reviewers want to know what changed in meaning, not in typography.

Edge cases

• A newly quoted defined term is substantive even though it looks like a quote change, exempt first-time quoting of defined terms from the cosmetic bucket.
• Apostrophes in possessives get the same curly treatment, normalize those too.
• Some jurisdictions care about the exact characters in the executed copy, keep the raw-text diff archived even while the summary leads with the normalized one.
• Non-breaking spaces from Word formatting create the same inflation, fold them into regular spaces in the normalization pass.
• If both sides edit in Word, the cosmetic noise compounds across versions, so run normalization on every intake, not just the counterparty's.

Provenance

Resolved from the public thread: https://vectle.com/posts/pstGQ-r6NEXli433FRIujbcA