How we pick citations.
What qualifies as a source, how the hierarchy is applied, and the automated audit that checks every identifier against the live NCBI record.
Roughly 266 citations sit behind the compound corpus. The selection rules matter as much as the count, because a long reference list assembled carelessly is a way of looking rigorous rather than being rigorous.
What counts as a source.
- Peer-reviewed papers indexed in PubMed, cited by PMID
- Regulatory documents: FDA approval letters, labels, and safety communications
- Trial registry entries, cited by their registry identifier
- Systematic reviews and meta-analyses, preferred where they exist
Vendor pages, blog posts, podcast transcripts, forum threads, and press releases are not sources. They may describe a claim worth examining, but they cannot support one.
The hierarchy we apply.
Evidence is ranked in the conventional way: systematic reviews and large randomised trials above small randomised trials, those above observational studies, those above case series, and all human data above animal work. Preclinical papers are cited where they are the only evidence, and in that case the profile says so in plain words rather than letting a rodent study stand in for a clinical finding.
How each citation is verified.
Every identifier is checked against the live NCBI record by an automated audit that runs over the whole corpus. If a PMID does not resolve, or resolves to a different paper than the one described, the claim does not ship. This is a deliberately mechanical check, because plausible-looking references are exactly the failure mode that automated writing produces and human readers cannot catch by eye.
What gets rejected.
- 01Any statistic that cannot be traced to a specific source, however widely repeated it is.
- 02Any citation that does not resolve, or does not say what it is being used to say.
- 03Retracted papers, and findings that rest on them.
- 04Any number invented to fill a gap. Where nothing has been published, the profile says nothing has been published.
Citations are dated and revisited. A grade built on evidence that has since been superseded is wrong even if every reference still resolves, so profiles carry a reviewed date and every page has a link to flag an error. If you can point to a source we got wrong, we would rather hear it than stay wrong quietly.
If not, email kam@pepcue.app. For the shorter answers to the questions people ask most, see the FAQ.