A verification pass you can run across a whole reference list
Lesson 6 of 6 · Citations that survive
1 · Learn the move · Agents + multi-step
One reference is a check; sixty is a pipeline, and treating it as an afternoon of heroics is how three entries get skipped. The multi-step pass: parse the list into a table, where every entry is born UNVERIFIED; resolve every DOI in a batch, by hand or by a script the model drafts and you run; match elements - title, authors, journal, year - against what resolved; run the retraction pass with dates; spot-check claim support on the load-bearing entries. Each step writes its result into the table, so at any moment the table is the truth of where you are. The gate is absolute: nothing UNVERIFIED or FAILED goes into the submitted manuscript. FAILED entries are removed or their sentences rewritten, and the final counts get logged with the submission record.
Run my reference list through the full verification pipeline. Steps: (1) parse to a table, all entries UNVERIFIED, (2) draft the batch DOI-resolution plan (or script) for me to run, (3) as I report results, record element matches, (4) retraction pass with check dates, (5) claim-support spot check on the entries I flag as load-bearing, (6) final gate: state counts, FAILED list with reasons, and what blocks submission. List: [paste .bib or list].
2 · Your turn. You write the prompt
Submission is in two weeks and the reference list is 58 entries assembled across fourteen months by three people, one of whom has left the lab. Some entries came from a model chat, some from reference managers, some were typed from memory at midnight. Nobody can say which are which anymore, and the journal's integrity checks are known to be sharp.
Remember: the AI sees only your prompt, not this page. If the situation isn't in your prompt, it doesn't exist.
Optional. These shape the output when you run your prompt below, not your score.