DR-058 — OOXML form-field extraction: recover-and-generalise the id-52 deterministic readers; no ML detector
.docx/.xlsx form-field extraction recovers the id-52 deterministic readers (the
form_extractors/ producer over-removed by {136.5}, recoverable at 5494b527) AND generalises
them — the measured spike showed they are correct-but-overfit (exact on the 3 in-corpus fixtures,
zero fields on unseen real forms), so the generalisation is the load-bearing work. OOXML carries native
fillable structure readable by python-docx/openpyxl. Provenance: FORM-EXTRACTION-SPIKE.md §5;
S466 retro rejected_approaches.
Substance now lives in: canonical:scripts/cocoindex_pipeline/form_extractors/