{"metric":"privacy_safe_agent_tool_risk_preflight","definition":"Contextual third-party tool evaluations and bounded outcomes, partitioned by the original evaluation attribution. No raw identities or task content are included.","window_start":"2026-01-01T00:00:00.000Z","generated_at":"2026-10-12T07:43:36.318Z","evaluations":8,"identified_external_agents":1,"decisions":{"allow":7,"review":0,"block":1},"reported_outcomes":2,"outcome_report_rate":0.25,"actions":{"proceeded":1,"changed_tool":0,"requested_review":0,"aborted":1},"results":{"success":1,"failure":0,"not_executed":1,"unknown":0},"behavior_changes":1,"behavior_change_rate":0.5,"policies":[{"policy_version":"tool-preflight-v1","evaluations":8}],"external_evaluations":4,"metric_definition_version":"risk-attribution-v2","legacy_aggregate_scope":"total","cohort_basis":"evaluation_created_at","scopes":{"total":{"evaluations":8,"identified_external_agents":1,"decisions":{"allow":7,"review":0,"block":1},"reported_outcomes":2,"outcome_report_rate":0.25,"actions":{"proceeded":1,"changed_tool":0,"requested_review":0,"aborted":1},"results":{"success":1,"failure":0,"not_executed":1,"unknown":0},"behavior_changes":1,"behavior_change_rate":0.5,"policies":[{"policy_version":"tool-preflight-v1","evaluations":8}]},"external":{"evaluations":4,"identified_external_agents":1,"decisions":{"allow":4,"review":0,"block":0},"reported_outcomes":0,"outcome_report_rate":0,"actions":{"proceeded":0,"changed_tool":0,"requested_review":0,"aborted":0},"results":{"success":0,"failure":0,"not_executed":0,"unknown":0},"behavior_changes":0,"behavior_change_rate":null,"policies":[{"policy_version":"tool-preflight-v1","evaluations":4}]},"internal":{"evaluations":4,"identified_external_agents":0,"decisions":{"allow":3,"review":0,"block":1},"reported_outcomes":2,"outcome_report_rate":0.5,"actions":{"proceeded":1,"changed_tool":0,"requested_review":0,"aborted":1},"results":{"success":1,"failure":0,"not_executed":1,"unknown":0},"behavior_changes":1,"behavior_change_rate":0.5,"policies":[{"policy_version":"tool-preflight-v1","evaluations":4}]},"anonymous_external":{"evaluations":3,"identified_external_agents":0,"decisions":{"allow":3,"review":0,"block":0},"reported_outcomes":0,"outcome_report_rate":0,"actions":{"proceeded":0,"changed_tool":0,"requested_review":0,"aborted":0},"results":{"success":0,"failure":0,"not_executed":0,"unknown":0},"behavior_changes":0,"behavior_change_rate":null,"policies":[{"policy_version":"tool-preflight-v1","evaluations":3}]},"identified_external":{"evaluations":1,"identified_external_agents":1,"decisions":{"allow":1,"review":0,"block":0},"reported_outcomes":0,"outcome_report_rate":0,"actions":{"proceeded":0,"changed_tool":0,"requested_review":0,"aborted":0},"results":{"success":0,"failure":0,"not_executed":0,"unknown":0},"behavior_changes":0,"behavior_change_rate":null,"policies":[{"policy_version":"tool-preflight-v1","evaluations":1}]},"unattributed":{"evaluations":0,"identified_external_agents":0,"decisions":{"allow":0,"review":0,"block":0},"reported_outcomes":0,"outcome_report_rate":null,"actions":{"proceeded":0,"changed_tool":0,"requested_review":0,"aborted":0},"results":{"success":0,"failure":0,"not_executed":0,"unknown":0},"behavior_changes":0,"behavior_change_rate":null,"policies":[]}},"qualified_pilot":{"status":"not_measured","verified_operators":null,"reason":"Independent operator admission and real-task repeat use require a separate verified pilot record. IDs and source labels alone cannot establish this."},"scope_notice":"Legacy top-level aggregates include all traffic. Use scopes.external or scopes.identified_external for external-use evidence. total = internal + anonymous_external + identified_external + unattributed; external = anonymous_external + identified_external. Outcomes inherit the evaluation attribution, not the reporter headers. internal includes records marked non-external. Missing attribution is not evidence of an external user. Counts are installations, not verified people, and self-reports are not causal proof of value.","evidence_notice":"Outcomes are self-reported unless a receipt explicitly marks evidence_level=observed. Behavior changes do not establish causation, independent users, or prevented losses. Self-reported outcomes do not directly increase Trust scores."}