How confidence is scored
Health reporting presents every finding with the same implied authority. This site doesn’t. Each thesis — every synthesis, comparison, and answered question — is graded on a transparent rubric, and every research source carries an evidence design label (randomised trial, cohort, meta-analysis, press release…) so you know how much weight a claim can bear before you read it.
The two halves of a score
Claude reads each thesis and the source pages it cites, then grades four dimensions — adversarially. Quotes are checked against sources; single-author dependencies and unsupported extrapolations cost points.
Computed mechanically from the wiki itself: how many sources the page cites, how many independent outlets they span, how recently it was updated. Fully reproducible.
The four judged dimensions
How well are the page's claims supported by the cited sources? 5 means every major claim is traceable to a source; 1 means mostly unsupported assertion.
Do multiple independent sources converge on the key claims? 5 means three or more independent authors/outlets agree; 1 means everything rests on a single voice.
Is the logic sound — are caveats, counterarguments, and the limits of the evidence acknowledged? 5 means the page attacks its own argument; 1 means it's one-sided.
How far do conclusions extrapolate beyond what the sources actually say? 5 stays within the evidence; 1 presents speculative leaps as fact.
Reading the levels
Multiply-sourced, claims verified against sources, self-critical. Safe to quote — with its stated caveats.
Well-evidenced with known, named gaps. Directionally reliable; check the rationale before leaning on specifics.
A promising frame resting on thin or single-author sourcing. Treat as informed opinion, not established fact.
Early thinking. A hypothesis filed so it can compound — not a position.
Honest limits
- The judge is an LLM. It verifies claims against the wiki’s own sources — not against the world. A page can score well and still be wrong if its sources are.
- Many sources here are journalism summarising studies. The evidence label on a source page names the study design actually behind it and the layer that was read (the paper, a press release, a newspaper); where the article overstates the study, the page says so and the study wins.
- A health thesis exists only once two independent sources bear on it. One trial, one lab or one outlet scores a 1–2 on triangulation by design.
- Nothing here is medical advice. Population, dose and duration stay attached to every claim precisely because a finding in 96 under-45s with obesity may not transfer to you.
- Scores are regenerated when the wiki changes (last: 2026-11-02). A low score isn’t a defect — it’s an honest reading of where a thesis currently stands.