{"n_tasks":38,"n_verifiable":38,"reported_success_rate":0.5,"verified_success_rate":0.4474,"silent_failure_rate":0.1579,"silent_failures":3,"false_alarm_rate":0.0526,"false_alarms":1,"brier":0.0986,"ece":0.065,"risk_coverage":[{"threshold":0.99,"coverage":0.1579,"risk":0.0,"n":3},{"threshold":0.95,"coverage":1.0,"risk":0.1579,"n":19}],"escalation":{"threshold":0.99,"coverage":0.1579,"risk":0.0,"n":3,"target_risk":0.02},"eval_ground_truth_pass_rate":0.675,"runs":40,"by_status":{"verified":17,"silent_failure":3,"unverified":0,"pending_approval":10,"failed":10,"killed":0,"running":0},"fleet":{"agents":5,"runs":40,"tasks":38,"tool_calls":0,"gate_blocks":0,"screened":0,"verified":17,"silent_failures":3,"false_alarms":1}}