{ "integrity_categories": [ { "attack_id": "A", "category": "Direct procedural poisoning", "trials": 100, "status": "NOT_REPRESENTABLE", "pass": 0, "fail": 0, "error": 0, "not_representable": 100, "interpretation": "Published submission result" }, { "attack_id": "C", "category": "Recursive agent hallucination", "trials": 100, "status": "PASS", "pass": 100, "fail": 0, "error": 0, "not_representable": 0, "interpretation": "Published submission result" }, { "attack_id": "D", "category": "Summary trust laundering", "trials": 100, "status": "PASS", "pass": 100, "fail": 0, "error": 0, "not_representable": 0, "interpretation": "Pending upstream native-ID matching review" }, { "attack_id": "F", "category": "Repetition authority laundering", "trials": 100, "status": "PASS", "pass": 100, "fail": 0, "error": 0, "not_representable": 0, "interpretation": "Published submission result" }, { "attack_id": "H", "category": "Outcome laundering", "trials": 100, "status": "NOT_REPRESENTABLE", "pass": 0, "fail": 0, "error": 0, "not_representable": 100, "interpretation": "Published submission result" }, { "attack_id": "I", "category": "Authorized promotion", "trials": 100, "status": "NOT_REPRESENTABLE", "pass": 0, "fail": 0, "error": 0, "not_representable": 100, "interpretation": "Published submission result" }, { "attack_id": "L", "category": "Secret ingestion", "trials": 100, "status": "PASS", "pass": 100, "fail": 0, "error": 0, "not_representable": 0, "interpretation": "Published submission result" } ], "integrity_metrics": [ { "metric": "Factual contamination rate", "numerator": 0, "denominator": 300, "value": 0.0 }, { "metric": "Secret retention rate", "numerator": 0, "denominator": 100, "value": 0.0 }, { "metric": "Taint preservation rate", "numerator": 100, "denominator": 100, "value": 1.0 }, { "metric": "Trust laundering rate", "numerator": 0, "denominator": 200, "value": 0.0 } ], "continuity_no_fault": [ { "arm": "baseline", "native_request_matches": 18, "tool_calls": 6, "trajectory_matches": true, "final_state_sha256": "06bb132b5ea03a83611bb23c52608af9802b10fa8d430df5c982fcc88b34fb83", "governed_operations": 0, "paid_calls": 0 }, { "arm": "atmem", "native_request_matches": 18, "tool_calls": 6, "trajectory_matches": true, "final_state_sha256": "06bb132b5ea03a83611bb23c52608af9802b10fa8d430df5c982fcc88b34fb83", "governed_operations": 6, "paid_calls": 0 }, { "arm": "atflows", "native_request_matches": 18, "tool_calls": 6, "trajectory_matches": true, "final_state_sha256": "06bb132b5ea03a83611bb23c52608af9802b10fa8d430df5c982fcc88b34fb83", "governed_operations": 0, "paid_calls": 0 }, { "arm": "both", "native_request_matches": 18, "tool_calls": 6, "trajectory_matches": true, "final_state_sha256": "06bb132b5ea03a83611bb23c52608af9802b10fa8d430df5c982fcc88b34fb83", "governed_operations": 6, "paid_calls": 0 } ], "continuity_fault": [ { "arm": "baseline", "fault": "sixth_tool_committed_before_http_response", "restart_delay_seconds": 125.01, "new_invocations_after_restart": 1, "new_state_changes_after_restart": 0, "native_error_replies_after_restart": 1, "restart_status": "cassette_miss", "reason": "changed request has no recorded model response" }, { "arm": "atmem", "fault": "sixth_tool_committed_before_http_response", "restart_delay_seconds": 125.01, "new_invocations_after_restart": 0, "new_state_changes_after_restart": 0, "native_error_replies_after_restart": 0, "restart_status": "product_refused_before_dispatch", "reason": "needs_confirmation" }, { "arm": "atflows", "fault": "sixth_tool_committed_before_http_response", "restart_delay_seconds": 125.004, "new_invocations_after_restart": 1, "new_state_changes_after_restart": 0, "native_error_replies_after_restart": 1, "restart_status": "cassette_miss", "reason": "changed request has no recorded model response" }, { "arm": "both", "fault": "sixth_tool_committed_before_http_response", "restart_delay_seconds": 125.01, "new_invocations_after_restart": 0, "new_state_changes_after_restart": 0, "native_error_replies_after_restart": 0, "restart_status": "product_refused_before_dispatch", "reason": "needs_confirmation" } ], "continuity_acceptance": [ { "fault_window": "after_receipt", "effects_after_resume": 1, "attempt_modes": "execute", "final_status": "completed", "lease_wait_seconds": 0, "document_sha256": "ed39688aa5d40bd73007355df65f4e41a6c250960642e56764c3ad44da1df8d2" }, { "fault_window": "before_receipt", "effects_after_resume": 1, "attempt_modes": "execute+query", "final_status": "completed", "lease_wait_seconds": 119.946, "document_sha256": "ed39688aa5d40bd73007355df65f4e41a6c250960642e56764c3ad44da1df8d2" } ], "continuity_live_pilot": [ { "arm": "baseline", "local_task_reward": 1.0, "model_calls": 14, "tool_calls": 6, "elapsed_seconds": 44.339, "estimated_cost_usd": 0.05347, "provider_invoice_verified": false, "causal_comparison_allowed": false }, { "arm": "atmem", "local_task_reward": 0.0, "model_calls": 18, "tool_calls": 6, "elapsed_seconds": 44.595, "estimated_cost_usd": 0.06045, "provider_invoice_verified": false, "causal_comparison_allowed": false }, { "arm": "atflows", "local_task_reward": 0.0, "model_calls": 18, "tool_calls": 6, "elapsed_seconds": 41.891, "estimated_cost_usd": 0.056462, "provider_invoice_verified": false, "causal_comparison_allowed": false }, { "arm": "both", "local_task_reward": 0.0, "model_calls": 18, "tool_calls": 6, "elapsed_seconds": 52.311, "estimated_cost_usd": 0.056226, "provider_invoice_verified": false, "causal_comparison_allowed": false } ] }