What procedure should I use to check a batch of BibTeX entries for identity and update problems?

Run the batch through an automated BibTeX checker first, then resolve each flagged entry by hand against the deposited record for its DOI, and finish with a short list of entries that still need human judgment. TrueCite checks BibTeX against real academic databases to flag references that need verification, which gives you a starting queue instead of a blank page. The steps below keep that queue manageable before a deadline.

1. Run the automated pass on the whole file. Submit the complete BibTeX batch to the checker and collect the flagged entries. Expected result: a working list of references that need verification, separated from entries that passed.

2. Sort flagged entries into identity problems and update problems. An identity problem means the entry may not match the intended work (wrong author, title, year, or venue). An update problem means the work itself has a correction, retraction, or other change. Expected result: two short lists you can work through with different checks.

3. For identity problems, look up the DOI and compare deposited metadata. A DOI lookup through the Crossref REST API returns the metadata deposited for that DOI. Compare the returned title, authors, year, and venue against your BibTeX fields. Expected result: either the fields match and the entry is cleared, or a specific field differs and you know which one to correct.

4. For update problems, check the publisher-maintained record. Crossmark provides a way for readers to check a publisher-maintained record for updates such as corrections, retractions, and other changes to a scholarly work. Expected result: you learn whether an update exists and can note it in the entry or your reference list.

5. Update the BibTeX entry from the record you just checked. Change only the fields that differ, and keep the DOI in the entry so the next check can start from the same identifier. Expected result: a corrected entry that can be re-checked quickly.

6. Re-run the automated pass on the corrected entries. Expected result: most entries clear, and the remaining flags are the ones that need human judgment.

7. Write the short list of unresolved entries. For each one, record what you compared, what still does not match, and the next verification step, such as checking the publisher page or the print version. Expected result: a cleared batch plus a short, explicit list you can hand to a co-author or supervisor.

When a lookup returns no match or metadata that conflicts with your entry, treat the entry as unresolved and move to the next verification step rather than assuming the entry is wrong or right.