Match confidence to evidence.
A narrow claim-auditing engine for X accounts. For each account we extract every health claim, map it to an evidence card on a 5-level truth hierarchy, and score how carefully the account communicates uncertainty, scope, counterevidence, and commercial incentive. One number per account, one grade, a per-claim receipt back to the exact post and an archive snapshot.
01Scored accounts
Accounts are pre-analyzed offline and cached here. Click a row for the per-claim breakdown, audit links, and evidence citations.
02What the score captures
Six components sum to a 100-point per-claim score. Claim scores aggregate into an account score via severity, reach, and recency weights — then two hard caps catch the worst failure modes.
- Evidence alignment. Does the strength of the post match the strength of the evidence card?
- Scoping quality. Does the post specify population, intervention, outcome — or generalize?
- Uncertainty discipline. Does the post use absolute language where evidence is weak?
- Counterevidence behavior. Does the post acknowledge conflicting evidence, or ignore it?
- Commercial integrity. Does the post sell the exact intervention being advocated?
- Safety discipline. Does the post flag risks proportional to the intervention?
03Thesis
Health Trust Score should measure whether an account matches its confidence, scope, and selling behavior to the actual strength of human evidence.
04Auditability
Every claim score links to the live post on X and a web.archive.org
snapshot. Every evidence card cites guidelines, Cochrane reviews, or
landmark trials — with type, url, and label.
If a card's grade or scope is wrong, the fix is visible in the evidence
corpus — not hidden in a model prompt.
05How to read the rest of the site
- Truth Hierarchy — H1–H5 evidence ladder, with worked examples.
- Algorithm — the six-component, 100-point claim score and how it aggregates.
- Methodology — how the evidence corpus is built, source hierarchy, how to challenge a score.
- Analyzer — scored accounts with full receipts + a request queue for new ones.