How to audit a scientific bibliography

Five references: duplicates, a wrong DOI and a retraction

This example combines public bibliographic records with deliberately introduced errors. The table is recalculated by the app engine using retained, dated Crossref metadata; a future live query may return different evidence.

Load the public example and run the local analysis. It produces five records, two in one duplicate group. The fifth lacks authors, year and DOI. Network requests start only when you select records and request verification. Download the BibTeX and a separate example of invalid DOI syntax to repeat the checks.

For the third record, accept only the DOI field after reviewing evidence. The score rises from 60 to 100 and three records now share the duplicate group. The original DOI and accepted change are retained; references are never removed automatically. Compare corrected-example.bib and corrected-audit.json with the input.

Weights: DOI 40%, title 30%, authors 15%, year 10%, journal 5%. Scores are normalized over comparable fields and are not probabilities that a paper is correct. Title and subtitle are compared together. No results, rate limits or provider failures leave a record unverified; they do not prove it false.

Crossref · 2026-09-24T23:01:13.882Z · REST API