verify-references

Verify these references: $ARGUMENTS

A citation that looks well-formed (real-sounding authors, a plausible journal, a DOI-shaped string) can still be wrong in three different ways: the paper doesn't exist at all (fabricated), the paper exists but the citation misstates it (wrong year, wrong journal, wrong authors), or the paper exists and is cited correctly but has since been retracted. Only a database lookup distinguishes these — don't judge a citation by how plausible it reads.

Process

1. Parse the input into individual entries

Figure out what you're checking before touching any tool:

• Single citation or claim ("Smith et al 2023 showed X, doi:10.1038/xyz") → one entry.
• .bib file → read it and parse each @article{key, title={...}, author={...}, year={...}, journal={...}, doi={...}} block into a record. Missing fields are fine — just note what wasn't stated.
• Pasted reference list (numbered [12] ..., or a References section from a paper draft) → split on the list's own numbering/formatting. Don't re-derive boundaries from prose; use whatever markers the list already uses.

For each entry, extract whatever is present: DOI, title, author(s), year, journal/venue. A DOI is the strongest anchor — resolve on that first when available. If two entries share the same DOI, resolve it once and reuse the record — don't repeat the same lookup.

2. Resolve each entry to an authoritative record

If a DOI is present:

tu run Crossref_get_work '{"doi":"10.1038/s41586-025-10014-0"}'

Crossref is the largest DOI registrar, not the only one — it covers most journal articles, but datasets, software, and many repository records (Zenodo, institutional archives) are registered with DataCite instead. A Crossref 404 on its own does not mean the DOI is fake; try DataCite before concluding that:

tu run DataCite_get_doi '{"doi":"10.5281/zenodo.1215979"}'

Only classify the DOI as fabricated/mistyped once both registries fail (also try stripping trailing punctuation first — a stray period or parenthesis from the citation's formatting is a common false negative). Cross-check the returned title/authors/year against one more independent source (openalex_get_work_by_doi or SemanticScholar_get_paper with {"paper_id":"DOI:<doi>"}) — registry metadata is occasionally sparse for older or unusual works.