Citation anomaly detectors

Generated by detect_citation_anomalies.py over 4467 review files, 24436 distinct PMIDs (24435 with a cached record).

Both checks produce candidates for adjudication, not verdicts.

Check A - unresolvable identifiers

1 cited PMIDs returned no PubMed record. An identifier that does not resolve cannot support any annotation.

PMID	Cited by
PMID:34521819	human/JAK1, human/STAT1, human/STAT2

Check B - paralog mismatch

A citation whose cached record names some members of a numbered symbol family but not a gene that cites it. Candidates for a citation copied across a family that in fact characterises one member.

(0 citation/family co-citation pairs were considered at >= 3 members; bare co-citation is not itself a defect and is dumped to family_clusters.tsv only.)

Not run (pass --check-paralogs). This check is documented as low precision: it misses alias-renamed cases and its hits are dominated by legitimate complex-wide papers. It is opt-in so its output is not mistaken for a findings list.