Verification API¶
Verification helpers evaluate learned policy profiles after training. They estimate satisfaction probabilities, unilateral deviation incentives, and AlphaRank summaries from rollouts.
Confidence Helpers¶
Satisfaction¶
Summarize per-seed satisfaction runs.
Interval endpoints are averaged only descriptively; certification is
decided per seed profile and aggregated as counts under certification.
Runs without stored thresholds (legacy reports) use
probability_thresholds or the default P_{>=0.5}.