What counts as a "real" citation database, and where the gaps are
"Verified against a real database" sounds like a solid guarantee until you ask which database, because the honest answer is: it depends, and no single one covers everything.
Semantic Scholar and OpenAlex are enormous and free to search, but they're built mostly from what's openly indexed — which means recent papers behind a strict publisher paywall, or older work that was never digitized with full metadata, can genuinely be missing even though the paper is real. CrossRef is closer to authoritative for DOIs specifically, since it's literally the registry DOIs come from, but it doesn't cover everything that lacks a DOI (a lot of older conference papers, for instance). arXiv is complete for what's on arXiv and only for what's on arXiv. Field-specific sources — DBLP for computer science, Europe PMC for biomedical work — are excellent within their field and useless outside it.
The practical result: a search across one database, even a big one, can come back empty for a paper that's real but just not in that particular index — and if you don't know that, "not found" starts looking like "doesn't exist," which is its own kind of false signal in the opposite direction from hallucination.
This is why checking against one database is a narrower net than checking against several at once and merging the results. Scout queries 10 of them in parallel — CrossRef, OpenAlex, Semantic Scholar, DBLP, Europe PMC, arXiv, DOAJ, HAL, IETF, and CORE — specifically so that a gap in one doesn't read as "this source is fake" when it's really just "this source isn't in this particular index." Every result also shows which specific databases it came from, so you can judge the coverage yourself instead of trusting a single black-box "verified" checkmark.