These numbers come straight from our database and our regression test runs. Nothing here is estimated.
1 labelled claims, custom set, model mock, run 9/27/2026.
Share of 385 verdicts where the evidence couldn't justify an answer, so we said so.
Total model spend divided by 432 completed checks.
Every completed check links to its sources.
We don't publish a corrections log yet. When a re-check changes a verdict, it will be listed here.