- Skip YYYYMMDD and YYMMDD date patterns at end of email
- Skip digit sequences longer than 4 characters
- Require non-digit before 4-digit years at end
- Add knid.nl/kabelnoord.nl to consumer domains (Friesland ISP)
- Add 11 missing regional archive domains to HERITAGE_DOMAIN_MAP
- Update recalculation script to re-extract email semantics
Results:
- 3,151 false birth years removed
- 'Likely wrong person' reduced from 533 to 325 (-39%)
- 2,944 candidates' scores boosted
- Updated `entity_review.py` to map email semantic fields from JSON.
- Expanded `email_semantics.py` with additional museum mappings.
- Introduced a new rule in `.opencode/rules/no-duplicate-ontology-mappings.md` to prevent duplicate ontology mappings.
- Added a backup JSON file for entity resolution candidates.
- Created `enrich_email_semantics.py` to enrich candidates with email semantic signals.
- Developed `merge_entity_reviews.py` to merge reviewed decisions from a backup into new candidates.
