H4 semantic-field review

The H4 packet asks, per dictionary and Amarakośa field, whether an apparent coverage anomaly is real or an artifact. Auto-triage mechanically resolves the rows it can prove; this page surfaces the rest for a human source-check (UC-RD / H4).

This Queue Proves

It proves the atlas can separate the mechanically-explainable rows (a "missing" lemma that is present under a looser headword fold, or an AP/AP90 delta that is a normalization variant) from the rows that genuinely need a scholar's source-read.

This Queue Does Not Prove

It does not decide any row. Auto-resolved rows are machine proposals with their evidence attached; a reviewer can override them, and nothing here is written to csl-orig.

Trust Block

Evidence: data/lexico/h4_semantic_field_review_packet.json (synced to src/data/lexico/), built from the M8 semantic-field artifacts and the dictionary headword sets.
Limitations: coverage is headword presence, not sense/citation coverage; the auto-resolve fold (slp1_form_key) is conservative; the covered-quality questions (thin entry, scope fit, reverse-index artifact) are judgement, not auto-resolvable.
Validation: generated by npm run build-h4-review-packet; auto-triage + this page by npm run build-review-worksheets-adjacent tooling; checked by npm test.
Owner repo: csl-atlas.
Next use: work the needs-review table below; the ordered markdown twin is docs/H4_REVIEW_WORKSHEET.md.

Auto-resolved = the lemma is present in the dictionary under a looser headword fold the strict key missed, so the "missing"/"delta" flag is a normalization variant — variant-headword (SKD) or normalization-risk (AP/AP90), each with its matched headword as evidence. The covered-quality sample types stay for human review by design.

Needs human review

Auto-resolved (machine-proposed, overridable)

Generated by npm run build-h4-review-packet; synced by npm run sync-site-data; validated by npm test. CC-BY-SA-4.0.