[xsvm]

Summary

docx validate rejected every document Word saved with a tracked equation edit:

word/document.xml: 8 errors
  Element '{…/wordprocessingml/2006/main}del': This element is not expected. (line 1)
  …

Word records a tracked edit inside an equation per math run: <m:r><w:del …><w:rPr>…</w:rPr><m:t>E</m:t></w:del></m:r> (one mark per run, plus one in each m:ctrlPr). ECMA-376's CT_R has no slot for w:ins/w:del, but Microsoft documents it as supported Word behavior: MS-OI29500, Part 1 §22.1.2.87 (e), "Word supports the w:ins and w:del elements inside a math run." (The m:ctrlPr marks are already valid.)

This also hits docx-cli's own edits. equations edit --track writes <w:del><m:oMath>…</m:oMath></w:del><w:ins>…</w:ins>, which validates; once a user opens and saves that file in Word, Word rewrites it into the per-run shape above and validate reports 6 errors. Reproduced with Word 365 on Windows (open + SaveAs2 via COM). The same false errors land in the raw gate's baseline.

The fix

validationXmlFor already preprocesses a clone (MCE strip) because real Word markup is valid OOXML the base schema doesn't know. It now also unwraps w:ins/w:del that sit directly inside an m:r, so what they wrap (m:rPr, w:rPr, m:t) is still validated against CT_R. The schemas stay as published, per src/core/raw/CLAUDE.md. Doc note added there.

Related: #27's fixture uses this same Word shape, so it now passes docx validate too.

Test plan

• bunx biome check ., bunx knip, bunx tsc --noEmit
• 2 new tests in tests/core/raw.test.ts: Word's w:del/w:ins inside a math run validate clean (failed before the fix); content inside the mark is still checked (a bogus element there is still reported)
• raw / validate test suites pass (136 tests)
• Two Word-saved files (a Word tracked deletion of two equations; a docx-cli tracked edit re-saved by Word) now report valid
• Full suite: CI on this PR

[kklimuk]

🤖 docx-cli triage

Thanks @xsvm, this is a clean, well-scoped fix. I reviewed head 30d47dd.

Claim: docx validate and the raw gate's schema check reported false errors for the per-math-run revision marks Word writes (<m:r><w:del>…<m:t>…</m:t></w:del></m:r>, MS-OI29500 Part 1 §22.1.2.87 (e)). The PR unwraps w:ins/w:del that sit directly inside an m:r, only in the copy that gets validated.

Before/after (minimal docx from create --from with $E=mc^2$, then the XML hand-edited):

case	main	this PR
Word shape: w:del + w:ins inside m:r	2 errors (del/ins not expected)	valid — 5 parts checked
w:del inside a regular w:r	1 error	1 error (still caught)
w:del directly in m:e wrapping a bare m:t	1 error	1 error (still caught)
content wrong inside the mark (m:t before m:rPr)	1 error (del)	1 error (m:rPr not expected, so the real problem is now named)
nested w:ins><w:del inside m:r	1 error	1 error (only the outer mark is unwrapped)
raw insert --xml '<w:p><m:oMath><m:r><w:del …><m:rPr>…</m:rPr><m:t>x</m:t></w:del></m:r></m:oMath></w:p>'	VALIDATION_FAILED (1 new error)	accepted as p1; the written document.xml keeps the <m:r><w:del …> untouched
raw insert of <w:r><w:del><w:t>	VALIDATION_FAILED	VALIDATION_FAILED

So the unwrap only touches the validation copy and only applies inside m:r. Wrong content inside the mark is still reported. Both sides of the raw gate's before/after comparison go through validationXmlFor (cli/raw/commit.ts, cli/raw/part.ts, cli/validate), so the fix affects them the same way.

Checks: bun run check ✅, bun run test:unit ✅ (1911 pass, 0 fail). /simplify single pass: no changes. I did not run /code-review this run, and I pushed nothing.

Docs: the src/core/raw/CLAUDE.md addition matches the code. Small optional nit: the docstring on validateWmlXml (validate.ts ~L43) still describes only the markup-compatibility clean-up and could mention the new unwrap too.

Decisions for the maintainer

• Do you want this at all? No issue asked for it. It fixes false validate errors on files Word saved with tracked equation edits, including docx-cli's own equations edit --track output after a Word re-save.
• Follow-up, outside this PR: the reader doesn't see these math-run marks either, on main or here. track-changes list says "no tracked changes" for the Word-shaped file. read renders $Ec^2$, dropping both the deleted = and the inserted m. A raw-inserted deleted run reads as a live $x$. Now that the raw gate accepts this shape, that's a gap in the "every change shows up on the next read" rule. Worth an issue?

Action needed: approve/merge — PR #31 is ready as-is (I pushed nothing, so your own approval satisfies the last-push rule). Please also decide on the two checklist items above.

---

Generated by Claude Code

[kklimuk]

@xsvm hold off from making further changes till i've had a bit of time to read through and do my own reviews, it's gonna take me a few days.