PaperTrail
A provenance and verification agent for clinical-trial efficacy claims.
Paste a claim like “Drug X reduced events by 30%.” PaperTrail finds the primary source in PubMed or ClinicalTrials.gov, extracts what the source actually found, and flags any discrepancy — with every flag anchored to the exact words in the source.
How it works
- Stage 1
Retrieval
The claim is embedded and matched against cached PubMed and ClinicalTrials.gov sources with pgvector similarity search. Sources are cached on first fetch, so verification never depends on live external-API latency.
- Stage 2
Extraction
Claude reads the matched source and extracts a structured finding — effect size, population, endpoint, and caveats — validated against a strict schema before use. Raw model JSON is never trusted unvalidated.
- Stage 3
Verification
Claude compares the claim to the extracted finding and pairs its verdict with a deterministic effect-size cross-check over parsed estimates and confidence intervals. The result is a discrepancy type, a trust score, and exact flagged spans.
Who it’s for
Translational researchers & postdocs
Drafting a grant progress report or manuscript, citing sources they may not have re-read recently, and personally accountable for every claim to reviewers, funders, and their PI.
Medical & regulatory reviewers
Checking that efficacy statements in publications and communications faithfully represent the underlying trial evidence, with an auditable citation trail behind every flag.
Deliberately narrow
PaperTrail checks clinical-trial efficacy claims against PubMed and ClinicalTrials.gov — not preprints, press releases, or arbitrary statements. A narrow tool that is fully correct is more useful to a researcher than a broad one that is only approximately right.