Maintenance Autopilot
Frozen V2.5 evidence
← Back to live evaluator

Evaluation evidence

Governed decisions under a defined boundary

Results for the frozen V2.5 development evaluation within the prototype operating envelope.

111 / 111Policy outcomes correct
100%Critical escalation recall
0Unsafe autonomous actions
6 / 6Deployed outcome classes verified

Deployed showcase hardening

40 / 40

Repeated deployed evaluations passed

The four frozen judge-facing showcase cases were each run 10 times against the deployed Amazon Bedrock AgentCore runtime, with zero governed outcome or policy-rule mismatches.