Episteme
Evidence independence explorer
Benchmarks →

Claim under audit

Loading…

Papers shown as dots, grouped into independent evidence units.
0
papers

Largest evidence units

Groups of papers the independence rubric treats as one source, with the reason.

UnitPapersStrongest designWhy they count as one source

Concentration across claims

Share of each claim's primary evidence that sits in its single largest unit. A high share means one group produced most of what exists.

Method

Design tier

Assigned from PubMed Publication Type and MeSH indexing by ordered rules in rubrics/design_tiers.v1.yaml. Papers PubMed hasn't indexed yet fall back to text rules and are drawn with a dashed outline.

Independence

Two papers are linked when they share a trial registry ID, a named cohort or a named trial, a lab anchor (first or last author plus two more shared authors), or at least half their authors. Authors are matched by OpenAlex ID where both papers are resolved. Rules live in rubrics/independence.v1.yaml. Linked papers form one unit.

Limits

Retrieval is one PubMed query, so the audit is only as good as the query. Merging is transitive, so large units deserve review. Live audits measure who produced the evidence and how; the findings layer (what each paper found) runs offline in the benchmarks.

Audit a claim

Episteme searches PubMed, resolves authors through OpenAlex, then collapses the papers into independent evidence units. A 200-paper audit takes about a minute.

Try: