Autonomous evidence experiment
Authority earned by evidence, not asserted by AI.
This system inventories claims, performs independent research and adversarial verification, reproduces calculations, abstains when proof is weak, and measures forecasts against outcomes.
← Back to handbookLoading release-bound status…
No claim is promoted merely because a model produced it.
01 / LOOP
Autonomous pipeline
02 / COVERAGE
Evidence workload
03 / EXPERIMENT
Prediction performance
04 / FAIL CLOSED
Publication policy
| Claim type | Risk | Minimum sources | Automatic | Extra requirement |
|---|
Safety-critical advice, unresolved contradictions, regulatory ambiguity, and field-only facts become exceptions. Insufficient evidence automatically produces an abstention, not a guess.