Policy
How we use AI
Which steps use AI models, which models, and what a person checks.
Updated 2026-10-05
Where AI helps
We use Claude models made by Anthropic for these steps:
- Drafting claim text, relations and plain-language summaries from the sources.
- Automated stress tests: AI agents with separate briefs look for errors, confounds and overstatement in a posed reading.
- Writing the code that builds this site and checks its content.
Where AI does not decide
AI is never credited as an author or a reviewer. An automated stress test is not peer review.
Nothing is published until the publisher approves the change. Counts on the trust panel are computed by code. Which studies to include, the tier of each claim and every posed reading are editorial judgments.
Human verification so far
Independent human expert review: 0 of 1 published synthesis.
Each posed reading says how it was tested and against what.