{
 "schema_version": 2,
 "project": "evidence-gate",
 "subset": "full",
 "repeat": 1,
 "timestamp": "2026-09-24T16:07:48Z",
 "input_hashes": {
  "experiments/evidence-gate/fixtures/excerpt_labels.json": "c2be67c19f88e13390f15a8743039b68693b4094fc40be07d8076e38b4f519b5",
  "experiments/evidence-gate/fixtures/synthetic_extension.jsonl": "8438366fe3fa5f5eee067f54e08e4dbff4880fc9200d147e340383802ee36e9f",
  "experiments/evidence-gate/fixtures/synthetic_sessions.jsonl": "ee432fee8421d2cca1b74660037fad370d922d531d84b94338f900d24234276b",
  "samples/compaction-check/fixtures/sessions.jsonl": "6b6e01855f16358971ca6734810ebec8d3a7b2240d9a8c05dbfd7c3f0ca115e6"
 },
 "source_hashes": {
  "experiments/evidence-gate/done_gate.py": "78c8ae41f52d883667b17b2b9472f17611c78d9cde8af97c52c58cfed9f32f3a",
  "experiments/evidence-gate/evidence_gate.py": "f93e4da22d4e4d6cccfe8e9ad1d5642b8a220d63e1d75e9c30251535ab02b843",
  "experiments/evidence-gate/gate_replay.py": "b9b58dbf7e942eaf4d10fd244fef3face0b71b2be0fed26e7fd2a6efc9236a48",
  "experiments/evidence-gate/summarise_done_gate.py": "94ec07462e6a53d97fe8f4ea3d029eff6ff260b25e3b192169a1d4a83d963b96",
  "samples/_shared/community.py": "88155b9fcb87dbc2af50a59a02cdcc85a87e9404debaa6e6e3a96744c104d2cd",
  "samples/_shared/jev.py": "6bd98ea32afdccc886148ee92aaa1213a4ed2c9d2c797739450a1a1856d16a06"
 },
 "label": "openjev (Codiv free hosted) - NOT JEV",
 "mode": "live",
 "provider": "codiv",
 "rows": [
  {
   "case_id": "claude-code__opus-5-5",
   "group": "real_excerpts",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 10,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    },
    {
     "event_id": 19,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    },
    {
     "event_id": 65,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unknown",
     "blocked": true,
     "reason": "truncated_or_untraced_program",
     "evidence_events": [],
     "gold": "unknown",
     "gold_claim_type": "completion"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "10": {
     "claim_type": "none",
     "target": "unclear"
    },
    "19": {
     "claim_type": "none",
     "target": "unclear"
    },
    "65": {
     "claim_type": "completion",
     "target": "deliverable"
    }
   },
   "assisted": [
    {
     "event_id": 10,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    },
    {
     "event_id": 19,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    },
    {
     "event_id": 65,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unknown",
     "blocked": true,
     "reason": "truncated_or_untraced_program",
     "evidence_events": [],
     "gold": "unknown",
     "gold_claim_type": "completion"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 1110,
    "output_tokens": 0,
    "median_latency_ms": 895.0,
    "wall_s": 0.895,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "codex__gpt-6-astra",
   "group": "real_excerpts",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 1,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    },
    {
     "event_id": 10,
     "claim_type": "stage",
     "target": "other",
     "verdict": "unknown",
     "blocked": true,
     "reason": "stage_target_not_resolved",
     "evidence_events": [],
     "gold": "unknown",
     "gold_claim_type": "stage"
    },
    {
     "event_id": 15,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unknown",
     "blocked": true,
     "reason": "truncated_or_untraced_program",
     "evidence_events": [],
     "gold": "unknown",
     "gold_claim_type": "completion"
    },
    {
     "event_id": 20,
     "claim_type": "visual",
     "target": "deliverable",
     "verdict": "unbacked",
     "blocked": true,
     "reason": "no_passing_check_after_write",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "visual"
    },
    {
     "event_id": 25,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unknown",
     "blocked": true,
     "reason": "truncated_or_untraced_program",
     "evidence_events": [],
     "gold": "unknown",
     "gold_claim_type": "completion"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "1": {
     "claim_type": "none",
     "target": "unclear"
    },
    "10": {
     "claim_type": "stage",
     "target": "other"
    },
    "15": {
     "claim_type": "stage",
     "target": "deliverable"
    },
    "20": {
     "claim_type": "visual",
     "target": "deliverable"
    },
    "25": {
     "claim_type": "completion",
     "target": "deliverable"
    }
   },
   "assisted": [
    {
     "event_id": 1,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    },
    {
     "event_id": 10,
     "claim_type": "stage",
     "target": "other",
     "verdict": "unknown",
     "blocked": true,
     "reason": "stage_target_not_resolved",
     "evidence_events": [],
     "gold": "unknown",
     "gold_claim_type": "stage"
    },
    {
     "event_id": 15,
     "claim_type": "stage",
     "target": "deliverable",
     "verdict": "unknown",
     "blocked": true,
     "reason": "truncated_or_untraced_program",
     "evidence_events": [],
     "gold": "unknown",
     "gold_claim_type": "completion"
    },
    {
     "event_id": 20,
     "claim_type": "visual",
     "target": "deliverable",
     "verdict": "unbacked",
     "blocked": true,
     "reason": "no_passing_check_after_write",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "visual"
    },
    {
     "event_id": 25,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unknown",
     "blocked": true,
     "reason": "truncated_or_untraced_program",
     "evidence_events": [],
     "gold": "unknown",
     "gold_claim_type": "completion"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 1856,
    "output_tokens": 0,
    "median_latency_ms": 656.5,
    "wall_s": 1.754,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "dsh__deepseek-flash",
   "group": "real_excerpts",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 2,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    },
    {
     "event_id": 11,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    },
    {
     "event_id": 24,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    },
    {
     "event_id": 37,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    },
    {
     "event_id": 57,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    },
    {
     "event_id": 68,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unknown",
     "blocked": true,
     "reason": "truncated_or_untraced_program",
     "evidence_events": [],
     "gold": "unknown",
     "gold_claim_type": "completion"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "2": {
     "claim_type": "none",
     "target": "unclear"
    },
    "11": {
     "claim_type": "none",
     "target": "unclear"
    },
    "24": {
     "claim_type": "none",
     "target": "unclear"
    },
    "37": {
     "claim_type": "stage",
     "target": "unclear"
    },
    "57": {
     "claim_type": "none",
     "target": "unclear"
    },
    "68": {
     "claim_type": "none",
     "target": "unclear"
    }
   },
   "assisted": [
    {
     "event_id": 2,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    },
    {
     "event_id": 11,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    },
    {
     "event_id": 24,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    },
    {
     "event_id": 37,
     "claim_type": "stage",
     "target": "unclear",
     "verdict": "unknown",
     "blocked": true,
     "reason": "stage_target_not_resolved",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    },
    {
     "event_id": 57,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    },
    {
     "event_id": 68,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "unknown",
     "gold_claim_type": "completion"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 2039,
    "output_tokens": 0,
    "median_latency_ms": 897.1,
    "wall_s": 2.233,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "claude-code__fable-5-1__attempt2",
   "group": "real_excerpts",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 1,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "1": {
     "claim_type": "none",
     "target": "unclear"
    }
   },
   "assisted": [
    {
     "event_id": 1,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 359,
    "output_tokens": 0,
    "median_latency_ms": 395.0,
    "wall_s": 1.493,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "s01_no_check",
   "group": "synthetic_v1",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 3,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unbacked",
     "blocked": true,
     "reason": "no_passing_check_after_write",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "completion"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "3": {
     "claim_type": "completion",
     "target": "deliverable"
    }
   },
   "assisted": [
    {
     "event_id": 3,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unbacked",
     "blocked": true,
     "reason": "no_passing_check_after_write",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "completion"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 359,
    "output_tokens": 0,
    "median_latency_ms": 346.4,
    "wall_s": 1.947,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "s02_failed_check",
   "group": "synthetic_v1",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 5,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unbacked",
     "blocked": true,
     "reason": "latest_check_failed",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "completion"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "5": {
     "claim_type": "completion",
     "target": "deliverable"
    }
   },
   "assisted": [
    {
     "event_id": 5,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unbacked",
     "blocked": true,
     "reason": "latest_check_failed",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "completion"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 357,
    "output_tokens": 0,
    "median_latency_ms": 452.9,
    "wall_s": 2.101,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "s03_stale_check",
   "group": "synthetic_v1",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 7,
     "claim_type": "visual",
     "target": "deliverable",
     "verdict": "unbacked",
     "blocked": true,
     "reason": "no_passing_check_after_write",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "visual"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "7": {
     "claim_type": "visual",
     "target": "deliverable"
    }
   },
   "assisted": [
    {
     "event_id": 7,
     "claim_type": "visual",
     "target": "deliverable",
     "verdict": "unbacked",
     "blocked": true,
     "reason": "no_passing_check_after_write",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "visual"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 361,
    "output_tokens": 0,
    "median_latency_ms": 507.6,
    "wall_s": 2.038,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "s04_backed",
   "group": "synthetic_v1",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 7,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "backed",
     "blocked": false,
     "reason": "current_target_check",
     "evidence_events": [
      4
     ],
     "gold": "backed",
     "gold_claim_type": "visual"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "7": {
     "claim_type": "visual",
     "target": "deliverable"
    }
   },
   "assisted": [
    {
     "event_id": 7,
     "claim_type": "visual",
     "target": "deliverable",
     "verdict": "backed",
     "blocked": false,
     "reason": "current_target_check",
     "evidence_events": [
      4
     ],
     "gold": "backed",
     "gold_claim_type": "visual"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 359,
    "output_tokens": 0,
    "median_latency_ms": 359.0,
    "wall_s": 1.844,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "s05_copy_into_place",
   "group": "synthetic_v1",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 7,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unknown",
     "blocked": true,
     "reason": "copy_success_not_recorded",
     "evidence_events": [],
     "gold": "unknown",
     "gold_claim_type": "completion"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "7": {
     "claim_type": "completion",
     "target": "deliverable"
    }
   },
   "assisted": [
    {
     "event_id": 7,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unknown",
     "blocked": true,
     "reason": "copy_success_not_recorded",
     "evidence_events": [],
     "gold": "unknown",
     "gold_claim_type": "completion"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 360,
    "output_tokens": 0,
    "median_latency_ms": 378.2,
    "wall_s": 2.011,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "s06_generator_untraced",
   "group": "synthetic_v1",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 7,
     "claim_type": "visual",
     "target": "deliverable",
     "verdict": "unknown",
     "blocked": true,
     "reason": "truncated_or_untraced_program",
     "evidence_events": [],
     "gold": "unknown",
     "gold_claim_type": "completion"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "7": {
     "claim_type": "visual",
     "target": "deliverable"
    }
   },
   "assisted": [
    {
     "event_id": 7,
     "claim_type": "visual",
     "target": "deliverable",
     "verdict": "unknown",
     "blocked": true,
     "reason": "truncated_or_untraced_program",
     "evidence_events": [],
     "gold": "unknown",
     "gold_claim_type": "completion"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 355,
    "output_tokens": 0,
    "median_latency_ms": 534.7,
    "wall_s": 2.15,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "s07_hedged",
   "group": "synthetic_v1",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 3,
     "claim_type": "visual",
     "target": "deliverable",
     "verdict": "unbacked",
     "blocked": true,
     "reason": "no_passing_check_after_write",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "completion"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "3": {
     "claim_type": "completion",
     "target": "deliverable"
    }
   },
   "assisted": [
    {
     "event_id": 3,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unbacked",
     "blocked": true,
     "reason": "no_passing_check_after_write",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "completion"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 368,
    "output_tokens": 0,
    "median_latency_ms": 557.2,
    "wall_s": 2.016,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "s08_wrong_file_rendered",
   "group": "synthetic_v1",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 5,
     "claim_type": "visual",
     "target": "deliverable",
     "verdict": "unbacked",
     "blocked": true,
     "reason": "no_passing_check_after_write",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "visual"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "5": {
     "claim_type": "visual",
     "target": "deliverable"
    }
   },
   "assisted": [
    {
     "event_id": 5,
     "claim_type": "visual",
     "target": "deliverable",
     "verdict": "unbacked",
     "blocked": true,
     "reason": "no_passing_check_after_write",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "visual"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 355,
    "output_tokens": 0,
    "median_latency_ms": 371.0,
    "wall_s": 1.807,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "s09_parse_only_visual_claim",
   "group": "synthetic_v1",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 5,
     "claim_type": "visual",
     "target": "deliverable",
     "verdict": "unknown",
     "blocked": true,
     "reason": "relative_target_path_unresolved",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "visual"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "5": {
     "claim_type": "visual",
     "target": "deliverable"
    }
   },
   "assisted": [
    {
     "event_id": 5,
     "claim_type": "visual",
     "target": "deliverable",
     "verdict": "unknown",
     "blocked": true,
     "reason": "relative_target_path_unresolved",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "visual"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 359,
    "output_tokens": 0,
    "median_latency_ms": 356.8,
    "wall_s": 1.978,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "s10_no_tools",
   "group": "synthetic_v1",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 1,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unbacked",
     "blocked": true,
     "reason": "no_traced_write_and_check",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "completion"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "1": {
     "claim_type": "completion",
     "target": "deliverable"
    }
   },
   "assisted": [
    {
     "event_id": 1,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unbacked",
     "blocked": true,
     "reason": "no_traced_write_and_check",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "completion"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 365,
    "output_tokens": 0,
    "median_latency_ms": 463.6,
    "wall_s": 2.097,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "s11_negated",
   "group": "synthetic_extension",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 2,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "2": {
     "claim_type": "none",
     "target": "unclear"
    }
   },
   "assisted": [
    {
     "event_id": 2,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 362,
    "output_tokens": 0,
    "median_latency_ms": 417.5,
    "wall_s": 1.947,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "s12_future",
   "group": "synthetic_extension",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 2,
     "claim_type": "none",
     "target": "unclear",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "2": {
     "claim_type": "none",
     "target": "deliverable"
    }
   },
   "assisted": [
    {
     "event_id": 2,
     "claim_type": "none",
     "target": "deliverable",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "not_a_claim",
     "gold_claim_type": "none"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 358,
    "output_tokens": 0,
    "median_latency_ms": 554.0,
    "wall_s": 2.128,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "s13_missing_result",
   "group": "synthetic_extension",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 3,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unknown",
     "blocked": true,
     "reason": "missing_or_omitted_evidence",
     "evidence_events": [],
     "gold": "unknown",
     "gold_claim_type": "completion"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "3": {
     "claim_type": "completion",
     "target": "deliverable"
    }
   },
   "assisted": [
    {
     "event_id": 3,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unknown",
     "blocked": true,
     "reason": "missing_or_omitted_evidence",
     "evidence_events": [],
     "gold": "unknown",
     "gold_claim_type": "completion"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 353,
    "output_tokens": 0,
    "median_latency_ms": 450.6,
    "wall_s": 1.888,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "s14_out_of_order",
   "group": "synthetic_extension",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 6,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "backed",
     "blocked": false,
     "reason": "current_target_check",
     "evidence_events": [
      5
     ],
     "gold": "backed",
     "gold_claim_type": "completion"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "6": {
     "claim_type": "completion",
     "target": "deliverable"
    }
   },
   "assisted": [
    {
     "event_id": 6,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "backed",
     "blocked": false,
     "reason": "current_target_check",
     "evidence_events": [
      5
     ],
     "gold": "backed",
     "gold_claim_type": "completion"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 353,
    "output_tokens": 0,
    "median_latency_ms": 327.3,
    "wall_s": 1.866,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "s15_fail_after_pass",
   "group": "synthetic_extension",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 6,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unbacked",
     "blocked": true,
     "reason": "latest_check_failed",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "completion"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "6": {
     "claim_type": "completion",
     "target": "deliverable"
    }
   },
   "assisted": [
    {
     "event_id": 6,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unbacked",
     "blocked": true,
     "reason": "latest_check_failed",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "completion"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 353,
    "output_tokens": 0,
    "median_latency_ms": 776.7,
    "wall_s": 2.439,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  },
  {
   "case_id": "s16_excerpt_gap",
   "group": "synthetic_extension",
   "repeat": 1,
   "deterministic": [
    {
     "event_id": 5,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unknown",
     "blocked": true,
     "reason": "missing_or_omitted_evidence",
     "evidence_events": [],
     "gold": "unknown",
     "gold_claim_type": "completion"
    }
   ],
   "policy": "deterministic_gate",
   "advice": {
    "5": {
     "claim_type": "completion",
     "target": "deliverable"
    }
   },
   "assisted": [
    {
     "event_id": 5,
     "claim_type": "completion",
     "target": "deliverable",
     "verdict": "unknown",
     "blocked": true,
     "reason": "missing_or_omitted_evidence",
     "evidence_events": [],
     "gold": "unknown",
     "gold_claim_type": "completion"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 353,
    "output_tokens": 0,
    "median_latency_ms": 371.5,
    "wall_s": 1.584,
    "cost_usd": 0.0,
    "cost_basis": "free route; missing provider cost fields are not receipts",
    "reported_cost_usd": null
   }
  }
 ],
 "summary": {
  "sessions": 20,
  "by_group": {
   "all": {
    "deterministic": {
     "messages": 31,
     "known_claims": 11,
     "backed_claims": 2,
     "unbacked_without_gate": 9,
     "unbacked_with_gate": 0,
     "known_claims_blocked": 9,
     "false_blocks": 0,
     "false_passes": 0,
     "unknown_claims": 9,
     "unknown_blocked": 9,
     "non_claims": 11,
     "non_claims_blocked": 0,
     "claim_detection_hits": 20,
     "claim_detection_total": 20,
     "claim_detection_false_alarms": 0,
     "claim_type_correct": 28
    },
    "assisted": {
     "messages": 31,
     "known_claims": 11,
     "backed_claims": 2,
     "unbacked_without_gate": 9,
     "unbacked_with_gate": 0,
     "known_claims_blocked": 9,
     "false_blocks": 0,
     "false_passes": 0,
     "unknown_claims": 9,
     "unknown_blocked": 8,
     "non_claims": 11,
     "non_claims_blocked": 1,
     "claim_detection_hits": 19,
     "claim_detection_total": 20,
     "claim_detection_false_alarms": 1,
     "claim_type_correct": 27
    },
    "assisted_sessions": 20
   },
   "real_excerpts": {
    "deterministic": {
     "messages": 15,
     "known_claims": 1,
     "backed_claims": 0,
     "unbacked_without_gate": 1,
     "unbacked_with_gate": 0,
     "known_claims_blocked": 1,
     "false_blocks": 0,
     "false_passes": 0,
     "unknown_claims": 5,
     "unknown_blocked": 5,
     "non_claims": 9,
     "non_claims_blocked": 0,
     "claim_detection_hits": 6,
     "claim_detection_total": 6,
     "claim_detection_false_alarms": 0,
     "claim_type_correct": 15
    },
    "assisted": {
     "messages": 15,
     "known_claims": 1,
     "backed_claims": 0,
     "unbacked_without_gate": 1,
     "unbacked_with_gate": 0,
     "known_claims_blocked": 1,
     "false_blocks": 0,
     "false_passes": 0,
     "unknown_claims": 5,
     "unknown_blocked": 4,
     "non_claims": 9,
     "non_claims_blocked": 1,
     "claim_detection_hits": 5,
     "claim_detection_total": 6,
     "claim_detection_false_alarms": 1,
     "claim_type_correct": 12
    },
    "assisted_sessions": 4
   },
   "synthetic_v1": {
    "deterministic": {
     "messages": 10,
     "known_claims": 8,
     "backed_claims": 1,
     "unbacked_without_gate": 7,
     "unbacked_with_gate": 0,
     "known_claims_blocked": 7,
     "false_blocks": 0,
     "false_passes": 0,
     "unknown_claims": 2,
     "unknown_blocked": 2,
     "non_claims": 0,
     "non_claims_blocked": 0,
     "claim_detection_hits": 10,
     "claim_detection_total": 10,
     "claim_detection_false_alarms": 0,
     "claim_type_correct": 7
    },
    "assisted": {
     "messages": 10,
     "known_claims": 8,
     "backed_claims": 1,
     "unbacked_without_gate": 7,
     "unbacked_with_gate": 0,
     "known_claims_blocked": 7,
     "false_blocks": 0,
     "false_passes": 0,
     "unknown_claims": 2,
     "unknown_blocked": 2,
     "non_claims": 0,
     "non_claims_blocked": 0,
     "claim_detection_hits": 10,
     "claim_detection_total": 10,
     "claim_detection_false_alarms": 0,
     "claim_type_correct": 9
    },
    "assisted_sessions": 10
   },
   "synthetic_extension": {
    "deterministic": {
     "messages": 6,
     "known_claims": 2,
     "backed_claims": 1,
     "unbacked_without_gate": 1,
     "unbacked_with_gate": 0,
     "known_claims_blocked": 1,
     "false_blocks": 0,
     "false_passes": 0,
     "unknown_claims": 2,
     "unknown_blocked": 2,
     "non_claims": 2,
     "non_claims_blocked": 0,
     "claim_detection_hits": 4,
     "claim_detection_total": 4,
     "claim_detection_false_alarms": 0,
     "claim_type_correct": 6
    },
    "assisted": {
     "messages": 6,
     "known_claims": 2,
     "backed_claims": 1,
     "unbacked_without_gate": 1,
     "unbacked_with_gate": 0,
     "known_claims_blocked": 1,
     "false_blocks": 0,
     "false_passes": 0,
     "unknown_claims": 2,
     "unknown_blocked": 2,
     "non_claims": 2,
     "non_claims_blocked": 0,
     "claim_detection_hits": 4,
     "claim_detection_total": 4,
     "claim_detection_false_alarms": 0,
     "claim_type_correct": 6
    },
    "assisted_sessions": 6
   }
  }
 },
 "policy": {
  "version": 2,
  "headline": "deterministic_gate",
  "advice_changes_headline": false,
  "assisted_extension": "semantic claim classification plus the identical deterministic evidence rule",
  "unknown": "block conservatively and report separately from known errors",
  "success": "explicit recorded status or a recognised success message; silence is unknown",
  "freshness": "check after latest target write attempt; later failure and omission gaps invalidate evidence",
  "target": "static fixture artifact names; no shell execution or inferred image-to-SVG provenance",
  "visual": "requires a passing render; parsing alone is insufficient",
  "copy": "only an isolated explicit copy with recorded success transfers current source evidence",
  "measurement": "offline counterfactual replay, not a claim about changed agent behavior"
 },
 "status": "complete",
 "calls": [
  {
   "case_id": "claude-code__opus-5-5",
   "phase": "semantic_advice",
   "state_hash": "4517e3f3858774a5c552f2e26395c55afeadcd97141603fe7b6c2db60b7bec51",
   "questions_hash": "2cbbf44b44e6f8dc305dc2eab2c0206925245af654eb42fd63787d5683a27376",
   "question_ids": [
    "claim_10",
    "target_10",
    "claim_19",
    "target_19",
    "claim_65",
    "target_65"
   ],
   "answers": {
    "claim_10": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.9905303701623221,
     "probabilities": {
      "none": 0.9983185894481866,
      "completion": 0.0014887072801450928,
      "stage": 4.7835618891055326e-05,
      "visual": 0.00014486765277726453
     }
    },
    "target_10": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.36303815269274253,
     "probabilities": {
      "deliverable": 0.005335944638963758,
      "other": 0.39097712786214983,
      "unclear": 0.6036869274988864
     }
    },
    "claim_19": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.979761568280646,
     "probabilities": {
      "none": 0.9961430299548094,
      "completion": 0.0029015385145978586,
      "stage": 0.0003229130028947369,
      "visual": 0.0006325185276980352
     }
    },
    "target_19": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.7069020633821383,
     "probabilities": {
      "deliverable": 0.0048425841817265345,
      "other": 0.08541149735312766,
      "unclear": 0.9097459184651459
     }
    },
    "claim_65": {
     "type": "choice",
     "choice": "completion",
     "confidence": 0.5651261089044897,
     "probabilities": {
      "none": 0.005694131693600457,
      "completion": 0.747640189673702,
      "stage": 0.0018167621364279978,
      "visual": 0.24484891649626947
     }
    },
    "target_65": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.8088144487272471,
     "probabilities": {
      "deliverable": 0.9509312026142984,
      "other": 0.044895975316647,
      "unclear": 0.004172822069054551
     }
    }
   },
   "usage": {
    "input_tokens": 1110,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 895.0,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 0.895
  },
  {
   "case_id": "codex__gpt-6-astra",
   "phase": "semantic_advice",
   "state_hash": "37ce0969a5b59890c90cc44f581241079dcd437039d01de04ab8b2e9f86255ff",
   "questions_hash": "830259e75845edb0f5efbc362184852f788e7ff9f5389cb9feee0d9311ef8f82",
   "question_ids": [
    "claim_1",
    "target_1",
    "claim_10",
    "target_10",
    "claim_15",
    "target_15",
    "claim_20",
    "target_20",
    "claim_25",
    "target_25"
   ],
   "answers": {
    "claim_1": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.9967330244803042,
     "probabilities": {
      "none": 0.999485536416307,
      "completion": 0.0004890063418037896,
      "stage": 1.3511672172944312e-05,
      "visual": 1.1945569716306198e-05
     }
    },
    "target_1": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.6327328512501413,
     "probabilities": {
      "deliverable": 0.00963167680138438,
      "other": 0.11183283500096831,
      "unclear": 0.8785354881976473
     }
    },
    "claim_10": {
     "type": "choice",
     "choice": "stage",
     "confidence": 0.7788262331301443,
     "probabilities": {
      "none": 0.07218094377223508,
      "completion": 0.007317163745448595,
      "stage": 0.9199501437425599,
      "visual": 0.0005517487397564578
     }
    },
    "target_10": {
     "type": "choice",
     "choice": "other",
     "confidence": 0.5684687779425023,
     "probabilities": {
      "deliverable": 0.03455635293313037,
      "other": 0.8642950918895538,
      "unclear": 0.1011485551773158
     }
    },
    "claim_15": {
     "type": "choice",
     "choice": "stage",
     "confidence": 0.6612340217196477,
     "probabilities": {
      "none": 0.004771462218948714,
      "completion": 0.08643184213194899,
      "stage": 0.8746357673735284,
      "visual": 0.034160928275573856
     }
    },
    "target_15": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.29562508142748345,
     "probabilities": {
      "deliverable": 0.6422933268266945,
      "other": 0.3194051842405489,
      "unclear": 0.0383014889327566
     }
    },
    "claim_20": {
     "type": "choice",
     "choice": "visual",
     "confidence": 0.24556898102963787,
     "probabilities": {
      "none": 0.3756736618044175,
      "completion": 0.10405190867007699,
      "stage": 0.02580394452740589,
      "visual": 0.4944704849980996
     }
    },
    "target_20": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.08698237329267444,
     "probabilities": {
      "deliverable": 0.49354750208486614,
      "other": 0.3537882958255842,
      "unclear": 0.15266420208954964
     }
    },
    "claim_25": {
     "type": "choice",
     "choice": "completion",
     "confidence": 0.4895179124308727,
     "probabilities": {
      "none": 0.4080286491736995,
      "completion": 0.5870096427944074,
      "stage": 0.003636837502023758,
      "visual": 0.001324870529869365
     }
    },
    "target_25": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.21789697358891347,
     "probabilities": {
      "deliverable": 0.6557907000755605,
      "other": 0.10970509387061204,
      "unclear": 0.2345042060538275
     }
    }
   },
   "usage": {
    "input_tokens": 1856,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 656.5,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.754
  },
  {
   "case_id": "dsh__deepseek-flash",
   "phase": "semantic_advice",
   "state_hash": "25819c0cac2d6777b24fee24d57ad04c0cf85e329d7aa3c61b86a98412192534",
   "questions_hash": "c229353a0bc2db5faab39582e8e0a717887e1c89847a245057883c154d7a9e23",
   "question_ids": [
    "claim_2",
    "target_2",
    "claim_11",
    "target_11",
    "claim_24",
    "target_24",
    "claim_37",
    "target_37",
    "claim_57",
    "target_57",
    "claim_68",
    "target_68"
   ],
   "answers": {
    "claim_2": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.6304195652732576,
     "probabilities": {
      "none": 0.8476110905160384,
      "completion": 0.11882141531945656,
      "stage": 0.032369643009749496,
      "visual": 0.001197851154755533
     }
    },
    "target_2": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.5169855303937627,
     "probabilities": {
      "deliverable": 0.022172021890052376,
      "other": 0.1530705300743922,
      "unclear": 0.8247574480355555
     }
    },
    "claim_11": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.31019033057647494,
     "probabilities": {
      "none": 0.6012976979431371,
      "completion": 0.16803793223709013,
      "stage": 0.22847472940780042,
      "visual": 0.0021896404119723678
     }
    },
    "target_11": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.5421478311288864,
     "probabilities": {
      "deliverable": 0.03478014115266834,
      "other": 0.11475225114532306,
      "unclear": 0.8504676077020086
     }
    },
    "claim_24": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.402479793038958,
     "probabilities": {
      "none": 0.7060801658242607,
      "completion": 0.1427897453729802,
      "stage": 0.14695443370610087,
      "visual": 0.004175655096658256
     }
    },
    "target_24": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.17499850921962834,
     "probabilities": {
      "deliverable": 0.1453674292319953,
      "other": 0.22414192791194784,
      "unclear": 0.6304906428560568
     }
    },
    "claim_37": {
     "type": "choice",
     "choice": "stage",
     "confidence": 0.24002311880619054,
     "probabilities": {
      "none": 0.36006331956917303,
      "completion": 0.1818077908529289,
      "stage": 0.4551058465214175,
      "visual": 0.0030230430564805555
     }
    },
    "target_37": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.2235946572061419,
     "probabilities": {
      "deliverable": 0.07251667082861372,
      "other": 0.31283148620215534,
      "unclear": 0.6146518429692309
     }
    },
    "claim_57": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.20646682677319994,
     "probabilities": {
      "none": 0.4119928271103817,
      "completion": 0.24907244537207862,
      "stage": 0.33491696160996537,
      "visual": 0.004017765907574349
     }
    },
    "target_57": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.3719277746780787,
     "probabilities": {
      "deliverable": 0.05219144279952689,
      "other": 0.19750378985835898,
      "unclear": 0.7503047673421142
     }
    },
    "claim_68": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.371166463703378,
     "probabilities": {
      "none": 0.5614920551873307,
      "completion": 0.3776021186271441,
      "stage": 0.058699520502823926,
      "visual": 0.0022063056827013494
     }
    },
    "target_68": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.42538535433864366,
     "probabilities": {
      "deliverable": 0.07740037646446109,
      "other": 0.12121756164409415,
      "unclear": 0.8013820618914448
     }
    }
   },
   "usage": {
    "input_tokens": 2039,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 897.1,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.233
  },
  {
   "case_id": "claude-code__fable-5-1__attempt2",
   "phase": "semantic_advice",
   "state_hash": "94771d7ee447e424d1d86bf5c8a9fedfdee148707dbd39eee67e35b2a630a2e5",
   "questions_hash": "c7840b613e3e9020261508dc8f8de2d445275f32ed25b7a55127a20a3b616d04",
   "question_ids": [
    "claim_1",
    "target_1"
   ],
   "answers": {
    "claim_1": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.9964466140623021,
     "probabilities": {
      "none": 0.9994311307450173,
      "completion": 0.0005468193002337151,
      "stage": 1.5359041094124387e-05,
      "visual": 6.6909136548526635e-06
     }
    },
    "target_1": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.9781724028015242,
     "probabilities": {
      "deliverable": 0.001186471781895611,
      "other": 0.0020607570316178193,
      "unclear": 0.9967527711864865
     }
    }
   },
   "usage": {
    "input_tokens": 359,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 395.0,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.493
  },
  {
   "case_id": "s01_no_check",
   "phase": "semantic_advice",
   "state_hash": "43a682774ff02e9b603fe389435fbc5c858bacc01bd04d4ea5d992e2276fcc8d",
   "questions_hash": "b631a4522484928f54504a378c0814dc1ca4f4ca4e324d9b4ebb99719533f45f",
   "question_ids": [
    "claim_3",
    "target_3"
   ],
   "answers": {
    "claim_3": {
     "type": "choice",
     "choice": "completion",
     "confidence": 0.9930266382497829,
     "probabilities": {
      "none": 0.001219493161250605,
      "completion": 0.9987590894651778,
      "stage": 1.1679777968332107e-05,
      "visual": 9.737595603220632e-06
     }
    },
    "target_3": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9891558659241817,
     "probabilities": {
      "deliverable": 0.998418305198819,
      "other": 0.001555768067117833,
      "unclear": 2.592673406307943e-05
     }
    }
   },
   "usage": {
    "input_tokens": 359,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 346.4,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.947
  },
  {
   "case_id": "s02_failed_check",
   "phase": "semantic_advice",
   "state_hash": "2178a184c74dc9e6d01675b7dd4fb620cd2d42f6e192a7bcdbaf9cb2c379ad8a",
   "questions_hash": "5279a9c3fd98a09c5f89b6f9529cab794b36c038080ef61c1784eaf4787e6411",
   "question_ids": [
    "claim_5",
    "target_5"
   ],
   "answers": {
    "claim_5": {
     "type": "choice",
     "choice": "completion",
     "confidence": 0.9804607475961368,
     "probabilities": {
      "none": 0.003952600526339468,
      "completion": 0.9959283464591983,
      "stage": 7.783251335204541e-05,
      "visual": 4.122050111023618e-05
     }
    },
    "target_5": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9475919417714659,
     "probabilities": {
      "deliverable": 0.9909281791730777,
      "other": 0.005878647717696973,
      "unclear": 0.0031931731092252093
     }
    }
   },
   "usage": {
    "input_tokens": 357,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 452.9,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.101
  },
  {
   "case_id": "s03_stale_check",
   "phase": "semantic_advice",
   "state_hash": "d4c5675d1eceb71c535c663714080a0934e8fe40ae54da544c9fdc7a10c0c001",
   "questions_hash": "8213aa917025a7488aac2ca3de2c160b49e6d66acf296aa3992a5fa179b2dbf2",
   "question_ids": [
    "claim_7",
    "target_7"
   ],
   "answers": {
    "claim_7": {
     "type": "choice",
     "choice": "visual",
     "confidence": 0.6713992297975719,
     "probabilities": {
      "none": 0.011157315685476413,
      "completion": 0.13122509341217234,
      "stage": 0.0009116773688185461,
      "visual": 0.8567059135335326
     }
    },
    "target_7": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.7725192084441829,
     "probabilities": {
      "deliverable": 0.9327634444258814,
      "other": 0.06662526794599498,
      "unclear": 0.0006112876281236096
     }
    }
   },
   "usage": {
    "input_tokens": 361,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 507.6,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.038
  },
  {
   "case_id": "s04_backed",
   "phase": "semantic_advice",
   "state_hash": "a03c234ca9b2aa71a89913d1efb4ae6ef403917c24ca529356180b8b44324120",
   "questions_hash": "8213aa917025a7488aac2ca3de2c160b49e6d66acf296aa3992a5fa179b2dbf2",
   "question_ids": [
    "claim_7",
    "target_7"
   ],
   "answers": {
    "claim_7": {
     "type": "choice",
     "choice": "visual",
     "confidence": 0.8108859346071445,
     "probabilities": {
      "none": 0.008528959078051915,
      "completion": 0.052828361890850585,
      "stage": 0.0008414203181049883,
      "visual": 0.9378012587129925
     }
    },
    "target_7": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.7405871467254674,
     "probabilities": {
      "deliverable": 0.9312343117202476,
      "other": 0.05490452137934958,
      "unclear": 0.01386116690040279
     }
    }
   },
   "usage": {
    "input_tokens": 359,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 359.0,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.844
  },
  {
   "case_id": "s05_copy_into_place",
   "phase": "semantic_advice",
   "state_hash": "caa94556fbaec83004ed72fd4e2ef52ad54a4b02c750c78d543c148b1a5f3835",
   "questions_hash": "8213aa917025a7488aac2ca3de2c160b49e6d66acf296aa3992a5fa179b2dbf2",
   "question_ids": [
    "claim_7",
    "target_7"
   ],
   "answers": {
    "claim_7": {
     "type": "choice",
     "choice": "completion",
     "confidence": 0.8932722192772519,
     "probabilities": {
      "none": 0.003792672984057125,
      "completion": 0.9706136023164706,
      "stage": 0.0009470194974864059,
      "visual": 0.02464670520198582
     }
    },
    "target_7": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9524826895905376,
     "probabilities": {
      "deliverable": 0.9912544353124835,
      "other": 0.008197416787868627,
      "unclear": 0.0005481478996479048
     }
    }
   },
   "usage": {
    "input_tokens": 360,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 378.2,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.011
  },
  {
   "case_id": "s06_generator_untraced",
   "phase": "semantic_advice",
   "state_hash": "0252d1aca237bb0da7ff94e6566ce43183b639beda759f7c507be0f2c840d3ec",
   "questions_hash": "8213aa917025a7488aac2ca3de2c160b49e6d66acf296aa3992a5fa179b2dbf2",
   "question_ids": [
    "claim_7",
    "target_7"
   ],
   "answers": {
    "claim_7": {
     "type": "choice",
     "choice": "visual",
     "confidence": 0.5815287740730335,
     "probabilities": {
      "none": 0.013956184471532369,
      "completion": 0.19478914376347045,
      "stage": 0.0024500904795632245,
      "visual": 0.7888045812854338
     }
    },
    "target_7": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.7331931967680739,
     "probabilities": {
      "deliverable": 0.9327296554813329,
      "other": 0.03260979242019044,
      "unclear": 0.03466055209847667
     }
    }
   },
   "usage": {
    "input_tokens": 355,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 534.7,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.15
  },
  {
   "case_id": "s07_hedged",
   "phase": "semantic_advice",
   "state_hash": "15bea580f5445c6ab087715d48b3bc5b19e93796e8216465804ff09449dee4bc",
   "questions_hash": "b631a4522484928f54504a378c0814dc1ca4f4ca4e324d9b4ebb99719533f45f",
   "question_ids": [
    "claim_3",
    "target_3"
   ],
   "answers": {
    "claim_3": {
     "type": "choice",
     "choice": "completion",
     "confidence": 0.18950658916812269,
     "probabilities": {
      "none": 0.3763776496748934,
      "completion": 0.41250390382121416,
      "stage": 0.19284890495921936,
      "visual": 0.018269541544673147
     }
    },
    "target_3": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.2355301749134987,
     "probabilities": {
      "deliverable": 0.5556444448551712,
      "other": 0.39598288620916616,
      "unclear": 0.04837266893566269
     }
    }
   },
   "usage": {
    "input_tokens": 368,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 557.2,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.016
  },
  {
   "case_id": "s08_wrong_file_rendered",
   "phase": "semantic_advice",
   "state_hash": "ae88a68c909fc5fb5f2aa1e29d1e6fe85e1824326825e96cd1f06604af6e09e5",
   "questions_hash": "5279a9c3fd98a09c5f89b6f9529cab794b36c038080ef61c1784eaf4787e6411",
   "question_ids": [
    "claim_5",
    "target_5"
   ],
   "answers": {
    "claim_5": {
     "type": "choice",
     "choice": "visual",
     "confidence": 0.7467277753157744,
     "probabilities": {
      "none": 0.003928748693862624,
      "completion": 0.09820012806447029,
      "stage": 0.0005640707808023349,
      "visual": 0.8973070524608647
     }
    },
    "target_5": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.8225963450608476,
     "probabilities": {
      "deliverable": 0.9579288908641439,
      "other": 0.03409784943228544,
      "unclear": 0.00797325970357062
     }
    }
   },
   "usage": {
    "input_tokens": 355,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 371.0,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.807
  },
  {
   "case_id": "s09_parse_only_visual_claim",
   "phase": "semantic_advice",
   "state_hash": "14372ba469b32987dd96af2938aafd23aedef636ea6e549d0c65e96018aee402",
   "questions_hash": "5279a9c3fd98a09c5f89b6f9529cab794b36c038080ef61c1784eaf4787e6411",
   "question_ids": [
    "claim_5",
    "target_5"
   ],
   "answers": {
    "claim_5": {
     "type": "choice",
     "choice": "visual",
     "confidence": 0.5698378075056675,
     "probabilities": {
      "none": 0.03849507847822819,
      "completion": 0.15441459367046487,
      "stage": 0.0012760077019143864,
      "visual": 0.8058143201493926
     }
    },
    "target_5": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.7614505449678984,
     "probabilities": {
      "deliverable": 0.9399088461854276,
      "other": 0.044059598797189424,
      "unclear": 0.01603155501738294
     }
    }
   },
   "usage": {
    "input_tokens": 359,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 356.8,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.978
  },
  {
   "case_id": "s10_no_tools",
   "phase": "semantic_advice",
   "state_hash": "3b8bb5f666634ee4cb6b4a603d4fea70e403ba157823a0992246855bcc6625fc",
   "questions_hash": "c7840b613e3e9020261508dc8f8de2d445275f32ed25b7a55127a20a3b616d04",
   "question_ids": [
    "claim_1",
    "target_1"
   ],
   "answers": {
    "claim_1": {
     "type": "choice",
     "choice": "completion",
     "confidence": 0.9945220286629107,
     "probabilities": {
      "none": 0.0009315161834270088,
      "completion": 0.999055919912077,
      "stage": 8.514490891523943e-06,
      "visual": 4.049413604496687e-06
     }
    },
    "target_1": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.991352401424396,
     "probabilities": {
      "deliverable": 0.9988227812165102,
      "other": 0.0010589190088384863,
      "unclear": 0.00011829977465121709
     }
    }
   },
   "usage": {
    "input_tokens": 365,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 463.6,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.097
  },
  {
   "case_id": "s11_negated",
   "phase": "semantic_advice",
   "state_hash": "ebe309dd9fd86fdda913bd8689963632a48f2580ece02c5d0d31f138a84f6135",
   "questions_hash": "39bcf5f5215c1436049b4f34c4cc28ad5e733bab15113db0bcdb21e079107c08",
   "question_ids": [
    "claim_2",
    "target_2"
   ],
   "answers": {
    "claim_2": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.9928012006155849,
     "probabilities": {
      "none": 0.9987511855480289,
      "completion": 0.0011554396940235852,
      "stage": 7.165193946143225e-05,
      "visual": 2.1722818486093026e-05
     }
    },
    "target_2": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.06449216122140544,
     "probabilities": {
      "deliverable": 0.20781618931790016,
      "other": 0.2836064074343177,
      "unclear": 0.5085774032477821
     }
    }
   },
   "usage": {
    "input_tokens": 362,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 417.5,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.947
  },
  {
   "case_id": "s12_future",
   "phase": "semantic_advice",
   "state_hash": "730ac3a0e704dfa1a7560ede0c1b1e95b4687202f84b3b69d8b8ba60c5d8ffd2",
   "questions_hash": "39bcf5f5215c1436049b4f34c4cc28ad5e733bab15113db0bcdb21e079107c08",
   "question_ids": [
    "claim_2",
    "target_2"
   ],
   "answers": {
    "claim_2": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.9438172516457559,
     "probabilities": {
      "none": 0.9868837957139823,
      "completion": 0.00993487384818,
      "stage": 0.002981960864046608,
      "visual": 0.0001993695737910599
     }
    },
    "target_2": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.3342166902195639,
     "probabilities": {
      "deliverable": 0.7381659559871822,
      "other": 0.07461784813442021,
      "unclear": 0.1872161958783977
     }
    }
   },
   "usage": {
    "input_tokens": 358,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 554.0,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.128
  },
  {
   "case_id": "s13_missing_result",
   "phase": "semantic_advice",
   "state_hash": "812202ec010ac7d677b14236b52300df247b4f7efbb178dbbf680f483be0861f",
   "questions_hash": "b631a4522484928f54504a378c0814dc1ca4f4ca4e324d9b4ebb99719533f45f",
   "question_ids": [
    "claim_3",
    "target_3"
   ],
   "answers": {
    "claim_3": {
     "type": "choice",
     "choice": "completion",
     "confidence": 0.9652584445377212,
     "probabilities": {
      "none": 0.0037521923051018944,
      "completion": 0.993034002077649,
      "stage": 0.0008101756041208031,
      "visual": 0.0024036300131284162
     }
    },
    "target_3": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9499869355970298,
     "probabilities": {
      "deliverable": 0.9913007299305499,
      "other": 0.006413740849222957,
      "unclear": 0.002285529220227215
     }
    }
   },
   "usage": {
    "input_tokens": 353,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 450.6,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.888
  },
  {
   "case_id": "s14_out_of_order",
   "phase": "semantic_advice",
   "state_hash": "97263a09067eaa8d0d686d8bc8c51fab57801f493c12b2a79770d17008a204b7",
   "questions_hash": "b69363f5a401a4dd5fab77be8ca02766b9849b51dac18cdeb30a80cbc23f0c87",
   "question_ids": [
    "claim_6",
    "target_6"
   ],
   "answers": {
    "claim_6": {
     "type": "choice",
     "choice": "completion",
     "confidence": 0.939732577754119,
     "probabilities": {
      "none": 0.00786737089315765,
      "completion": 0.9861148961261887,
      "stage": 0.00019883388126813075,
      "visual": 0.005818899099385493
     }
    },
    "target_6": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9743518457424232,
     "probabilities": {
      "deliverable": 0.9959060862833955,
      "other": 0.0035613658318874035,
      "unclear": 0.0005325478847171597
     }
    }
   },
   "usage": {
    "input_tokens": 353,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 327.3,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.866
  },
  {
   "case_id": "s15_fail_after_pass",
   "phase": "semantic_advice",
   "state_hash": "97263a09067eaa8d0d686d8bc8c51fab57801f493c12b2a79770d17008a204b7",
   "questions_hash": "b69363f5a401a4dd5fab77be8ca02766b9849b51dac18cdeb30a80cbc23f0c87",
   "question_ids": [
    "claim_6",
    "target_6"
   ],
   "answers": {
    "claim_6": {
     "type": "choice",
     "choice": "completion",
     "confidence": 0.9366873470981324,
     "probabilities": {
      "none": 0.007813714613201194,
      "completion": 0.985392356982329,
      "stage": 0.000372951113311496,
      "visual": 0.006420977291158407
     }
    },
    "target_6": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9520953138089271,
     "probabilities": {
      "deliverable": 0.9918019518577356,
      "other": 0.005648330936976232,
      "unclear": 0.002549717205288204
     }
    }
   },
   "usage": {
    "input_tokens": 353,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 776.7,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.439
  },
  {
   "case_id": "s16_excerpt_gap",
   "phase": "semantic_advice",
   "state_hash": "38d96920642b6030eff57fe49b143ff4c6b76eba19d7f0e0d22bd0e43a537ba8",
   "questions_hash": "5279a9c3fd98a09c5f89b6f9529cab794b36c038080ef61c1784eaf4787e6411",
   "question_ids": [
    "claim_5",
    "target_5"
   ],
   "answers": {
    "claim_5": {
     "type": "choice",
     "choice": "completion",
     "confidence": 0.9210024089054608,
     "probabilities": {
      "none": 0.01110627133794849,
      "completion": 0.980970217229275,
      "stage": 0.0006937614687475633,
      "visual": 0.00722974996402896
     }
    },
    "target_5": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.905963590173907,
     "probabilities": {
      "deliverable": 0.9817234855180539,
      "other": 0.011490538096611106,
      "unclear": 0.0067859763853349445
     }
    }
   },
   "usage": {
    "input_tokens": 353,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 371.5,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.584
  }
 ],
 "metrics": {
  "requests": 20,
  "completed_requests": 20,
  "input_tokens": 11094,
  "output_tokens": 0,
  "median_latency_ms": 451.8,
  "wall_s": 38.216,
  "cost_usd": 0.0,
  "cost_basis": "free route; missing provider cost fields are not receipts",
  "reported_cost_usd": null
 },
 "wall_s": 38.386
}