Checks whether an answer's claims are actually supported by the papers it cites — grounded in each paper's real results, with retractions flagged and overstatements caught. Runs entirely in your browser with your own Anthropic API key.
Stored only in your browser (localStorage). It's sent directly from your browser to Anthropic and to us never. The judge step needs it; link checks and retraction detection don't.
Draft claims & find sources turns your question into a list of claims, each with up to two real papers discovered across OpenAlex, Crossref & Europe PMC in parallel — then a citation-trail round that pulls in papers co-cited by the strongest hits for higher recall — deduped by DOI, retractions dropped, and ranked by trust signals (co-citation, citations, recency, open-access, evidence tier) (needs your key). Each claim is checked against all its references, so a claim backed by several sources reads stronger than a single-source one. Then review and Verify. Or scope first with Clarify, or add claims by hand below. (Tip: prefix a search with !oa, !crossref, !epmc or !pubmed to scope it to one provider.)
Drafted or added here — edit anything before verifying.
The self-improvement loop runs on every claim: it searches OpenAlex + Crossref + Europe PMC for 2 more corroborating papers and verifies the claim against each, and (if the claim is OVERSTATED/UNSUPPORTED) rewrites it to match the evidence — re-verifying until the claim is supported by enough independent sources, or rounds run out. Found papers are marked “🔎 found by loop” and feed the Worth-your-read list; tightened claims replace the originals above. (Needs your key; more rounds = more API calls.)
Paste a paper's DOI / PMID / PMCID. We pull its introduction and, for each sentence that cites a reference, check whether the cited paper actually supports it — using the paper's real inline citations — then show the intro with every claim highlighted by verdict. Open-access papers in Europe PMC only (needs your key).
Paste a claim whose citation is weak, wrong, or missing. We search across OpenAlex, Crossref, Europe PMC & Semantic Scholar (two query angles, deduped) and verify the claim against each candidate, returning real papers that actually support it — so you can replace a bad citation, not just flag it. (Needs your key; it checks several papers, so give it a moment.)