v2.4.2

Features

• detect multilingual clinical identifiers (#384)
• benchmark: add native German PII model lane (#382)
• detect names in structured party fields (#377)

Bug Fixes

• scope name stopword exclusions by language (#392)
• scope contextual address boundaries by language (#391)
• close clinical identifier boundary gaps (#386)
• scan single-quoted TypeScript vocabulary (#389)
• stop address expansion at a trailing coordinating conjunction (#372)
• anonymize-wasm: load WASI binding under Bun as well as Node (#385)
• detect missing English counsel given names (#383)
• detect Unicode organizations with short legal forms (#381)
• drop birth-number prose false positives (#367)

Performance

• index legal-form list suffixes (#393)

Refactor

• migrate name/unit false-positive vocabularies to data (#390)
• migrate assembler seed vocabularies to per-language data (#388)

Documentation

• publish benchmark results and clarify surfaces (#394)

Miscellaneous

• version packages (#376)
• release anonymize data independently (#396)
• reuse shared Rust lint config from ai-shared (#374)
• ban hardcoded vocabulary in source with a lint gate (#373)

bench

• add German legal NER evaluation (#369)