Miscitation review — supporting material
Supporting material for the Miscitation Review project.
Contents
aggregate_miscitations.py— reproducible aggregator. Walks every
genes/*/*/*-ai-review.yaml, pulls out eachreferences[].reference_review
block, and writes the register and the TSV below. It reads only what reviewers
have already recorded; it makes no judgment of its own.miscitation-register.md— generated register (do not edit by hand):
counts bycorrectness, counts by organism, and a table of every non-VERIFIED
reference with gene, identifier, title andreview_notes.reports/miscitations.tsv(written at the repo root, not here) — one row per
adjudicated reference, with the untruncatedreview_notes, for querying
and triage.
Regenerate
uv run python projects/MISCITATIONS/aggregate_miscitations.py
Re-run after new reviews land, or after adjudicating references in an existing
review. The register is regenerated wholesale from the current state of the YAML,
so it never needs hand-editing.
What counts as "flagged"
ReferenceCorrectnessEnum has six values. The register treats them in three groups:
| Group | Values | Meaning |
|---|---|---|
| Clean | VERIFIED |
Checked; identifier resolves to the intended, supporting paper |
| Unchecked | UNVERIFIED |
Adjudication started but this reference not yet checked |
| Flagged | WRONG_IDENTIFIER, MISCITED, DISPUTED, LOW_QUALITY |
A reviewer has positively judged this citation to be a problem |
Percentages on the project page are flagged / adjudicated, not flagged / all
references — most references in the repo carry no reference_review at all and
are simply unexamined.