openspec: text-extraction-office-completeness scaffolding

[rjzondervan]

Summary

• Scaffolds the openspec change text-extraction-office-completeness — a single OfficeDocumentWalker driving both extraction and anonymisation over the same content surface.
• Extends current 2-levels-deep walker to traverse tables (recursive), lists (ListItemRun), headers (per section, per variant), footers (per section, per variant), footnotes, endnotes, and text frames.
• Adds ODT as a first-class supported format alongside DOCX on both paths.

What's in this PR

Pure openspec scaffolding. Four artifacts:

• proposal.md — gap analysis (the two-level depth, missing structures, ODT-unsupported reality), pipeline shape, capability scope
• design.md — 10 decisions: single walker class for both read+mutate (D1), element visitation rules (D2), output ordering with section markers (D3), ODT integration via PhpWord's ODText reader/writer (D4), strtr longest-match-first semantics (D5), ADR-005 logging (D6), backwards-compat-for-DOCX-consumers (D7), forceReExtract opt-in for stale records (D8), extractWord → extractOfficeDocument rename with one-cycle alias (D9), fixture strategy (D10)
• specs/text-extraction-office-completeness/spec.md — 9 ADDED Requirements with scenarios: walker coverage, extractText shape, replace mutation, ODT extraction, ODT writer-back, DOCX entity substitution in all walker-covered structures, ADR-005 logging, pre-change DOCX superset guarantee, BLOCKING reopen-clean validation gate
• tasks.md — 10 sections from the walker class through TextExtractionService + DocumentProcessingHandler refactors, ODT-specific writer-back path, fixtures, unit + integration tests, the manual Word/LibreOffice validation gate (BLOCKING), docs, and quality gates

openspec validate text-extraction-office-completeness — clean.

Why now

Two specific bugs surfaced in recent operator testing on Woo dossiers:

1. Entity detection misses names in document headers, footers, and footnotes (case numbers and behandelaar identifiers escape redaction).
2. ODT inputs hit a replaceWordsInTextDocument path that does str_ireplace on the binary ZIP container — corrupts the file. Operators see "Anonymisation succeeded" then can't open the result.

This change closes both. Per design D7, DOCX extraction is strictly additive — no regression for existing consumers; ODT goes from broken to working.

Composition

Pairs with sister office-document-sanitization (separate PR #1589). Sanitiser strips wrappers; walker traverses surviving content. Both target DOCX + ODT. Can land independently or together.

Status

Ready for team review. No implementation has started — /opsx:apply runs only after this PR merges.

Test plan

• Review the 9 ADDED Requirements for completeness vs. the actual gap
• Confirm output ordering decisions (D3) make sense for downstream NER context
• Confirm extractWord deprecation window (D9) — one cycle, or longer?
• Confirm capability name text-extraction-office-completeness is acceptable

[github-actions]

Quality Report — ConductionNL/openregister @ 579196d

Check	PHP	Vue	Security	License	Tests
lint	✅
phpcs	✅
phpmd	✅
psalm	✅
phpstan	✅
phpmetrics	✅
eslint		✅
stylelint		✅
composer			✅	✅ 162/162
npm			✅	✅ 602/602
PHPUnit					✅
Newman					⏭️
Playwright					⏭️

---

Quality workflow — 2026-05-19 09:29 UTC

Download the full PDF report from the workflow artifacts.