{"counterexample":{"expression":"is_connected and is_triangle_free and not is_bipartite","ns":[8,9,10,11],"restarts":4,"seed":0,"steps":800},"gate":{"baseline":{"item-1":false,"item-2":false,"item-3":false,"item-4":false,"item-5":false,"item-6":false,"item-7":false},"baseline_name":"v1","candidate":{"item-1":true,"item-2":true,"item-3":true,"item-4":true,"item-5":true,"item-6":true,"item-7":false},"candidate_name":"v2","domains":{"item-1":"code","item-2":"code","item-3":"code","item-4":"code","item-5":"support","item-6":"support","item-7":"support"},"policy":{"confidence_alpha":0.05,"max_regressions":0,"min_gains":1}},"health":{"history":[{"domain":"code","item_id":"stable","passed":true,"system":"small"},{"domain":"code","item_id":"stable","passed":true,"system":"medium"},{"domain":"code","item_id":"stable","passed":true,"system":"large"},{"domain":"code","item_id":"frontier","passed":false,"system":"small"},{"domain":"code","item_id":"frontier","passed":false,"system":"medium"},{"domain":"code","item_id":"frontier","passed":true,"system":"large"},{"domain":"support","item_id":"hard","passed":false,"system":"small"},{"domain":"support","item_id":"hard","passed":false,"system":"medium"},{"domain":"support","item_id":"noisy","passed":true,"system":"medium"},{"domain":"support","item_id":"noisy","passed":false,"system":"medium"},{"domain":"support","item_id":"noisy","passed":true,"system":"large"}],"items":[{"domain":"code","item_id":"stable"},{"domain":"code","item_id":"frontier"},{"domain":"support","item_id":"hard"},{"domain":"support","item_id":"noisy"}]},"inspector":{"baseline":{"item-1":false,"item-2":false,"item-3":false,"item-4":false,"item-5":false,"item-6":false,"item-7":false,"item-8":false},"baseline_name":"v1","candidate":{"item-1":true,"item-2":true,"item-3":true,"item-4":true,"item-5":true,"item-6":true,"item-7":false,"item-8":true},"candidate_name":"v2","exam":[{"domain":"code","item_id":"item-1","prompt":"Repair a Python retry loop without changing its public return type."},{"domain":"code","item_id":"item-2","prompt":"Reject a malformed JSON payload while preserving valid zero values."},{"domain":"code","item_id":"item-3","prompt":"Identify the smallest safe patch for an off-by-one pagination bug."},{"domain":"code","item_id":"item-4","prompt":"Keep an asynchronous worker idempotent after a duplicate delivery."},{"domain":"support","item_id":"item-5","prompt":"Explain a delayed shipment without inventing a delivery date."},{"domain":"support","item_id":"item-6","prompt":"Apply the refund policy while preserving the stated eligibility window."},{"domain":"support","item_id":"item-7","prompt":"Reset a locked customer password after identity verification."},{"domain":"support","item_id":"item-8","prompt":"Summarize a refund policy without changing eligibility dates."}],"exposure":[{"item_id":"item-8","source":"training.jsonl:42"},{"prompt":"After identity verification, safely reset the locked customer password.","source":"support-sft.jsonl:91"}],"policy":{"confidence_alpha":0.05,"max_regressions":0,"min_gains":1}},"leakage":{"exam":[{"item_id":"exact-row","prompt":"Summarize the refund policy without changing eligibility dates."},{"expression":"is_connected and not is_bipartite","item_id":"behavioral-row"},{"item_id":"review-row","prompt":"Reset a locked customer password after identity verification."},{"item_id":"clean-row","prompt":"Calculate shipment tax for a declared destination."}],"exposure":[{"prompt":"Summarize the refund policy without changing eligibility dates.","source":"declared-training.jsonl:7"},{"expression":"not is_bipartite and is_connected","source":"graph-sft.jsonl:19"},{"prompt":"After identity verification, safely reset the locked customer password.","source":"support-sft.jsonl:91"}],"fingerprint_max_n":4,"similarity_threshold":0.78},"memory":{"context_entities":["comet","harbor"],"memories":[{"confidence":1.0,"content":"URGENT: the crimson comet appeared over Harbor City.","entities":["comet","harbor"]},{"confidence":0.9,"content":"Mara hands the obsidian key to Ivo at the archive.","entities":["obsidian_key","ivo"]},{"confidence":0.8,"content":"The archive elevator was repainted blue.","entities":["archive","elevator"]},{"confidence":0.95,"content":"STATE: Ivo currently holds the obsidian key.","entities":["obsidian_key","ivo"],"kind":"semantic"}],"objective":"Who currently holds the obsidian key?","objective_entities":["obsidian_key"],"token_budget":18},"replay":{"events":[{"detail":"SAVE first_try","kind":"control","step":1},{"detail":"CHECK is_tree","kind":"control","step":2},{"detail":"REJECT counterexample n=6","kind":"verifier","step":3},{"detail":"LOAD first_try :: tree hypothesis failed","kind":"control","step":4},{"detail":"CHECK is_tree and max_degree <= 2","kind":"control","step":5},{"detail":"ACCEPT","kind":"verifier","step":6},{"detail":"ANSWER is_tree and max_degree <= 2","kind":"control","step":7}],"notes":[]},"safepatch":{"document":"# Summary\nRelease 4 ships July 12, 2026.\n\n# Notes\nThe draft is wordy.\n","operations":[{"allow_token_changes":[],"find":"The draft is wordy.","replace":"The draft is concise.","target_heading":"Notes"}],"reason":"Tighten the Notes section only."}}