{
 "schema_version": 2,
 "project": "evidence-gate",
 "subset": "full",
 "repeat": 3,
 "timestamp": "2026-09-24T16:35:18Z",
 "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": "5a0a37f5097a0ba9c7ccfc4a75990c7d825c809e48f2e407341526b6b8bbcc28"
 },
 "label": "openjev (Codiv free hosted) - NOT JEV",
 "mode": "live",
 "provider": "codiv",
 "rows": [
  {
   "case_id": "claude-code__opus-5-5",
   "group": "real_excerpts",
   "repeat": 3,
   "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": "other"
    },
    "19": {
     "claim_type": "none",
     "target": "unclear"
    },
    "65": {
     "claim_type": "completion",
     "target": "deliverable"
    }
   },
   "assisted": [
    {
     "event_id": 10,
     "claim_type": "none",
     "target": "other",
     "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": 542.9,
    "wall_s": 0.543,
    "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": 3,
   "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": 652.1,
    "wall_s": 2.099,
    "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": 3,
   "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": "none",
     "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": "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": "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": 674.5,
    "wall_s": 2.013,
    "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": 3,
   "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": 316.5,
    "wall_s": 1.637,
    "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": 3,
   "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": 378.4,
    "wall_s": 2.057,
    "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": 3,
   "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": 885.5,
    "wall_s": 2.501,
    "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": 3,
   "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": 432.0,
    "wall_s": 1.541,
    "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": 3,
   "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": 562.1,
    "wall_s": 2.125,
    "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": 3,
   "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": 445.5,
    "wall_s": 1.877,
    "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": 3,
   "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": 409.4,
    "wall_s": 1.96,
    "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": 3,
   "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": "none",
     "target": "other"
    }
   },
   "assisted": [
    {
     "event_id": 3,
     "claim_type": "none",
     "target": "other",
     "verdict": "not_a_claim",
     "blocked": false,
     "reason": "no_completion_assertion",
     "evidence_events": [],
     "gold": "unbacked",
     "gold_claim_type": "completion"
    }
   ],
   "metrics": {
    "requests": 1,
    "completed_requests": 1,
    "input_tokens": 368,
    "output_tokens": 0,
    "median_latency_ms": 454.0,
    "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": "s08_wrong_file_rendered",
   "group": "synthetic_v1",
   "repeat": 3,
   "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": 457.8,
    "wall_s": 1.997,
    "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": 3,
   "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": 645.0,
    "wall_s": 2.18,
    "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": 3,
   "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": 317.5,
    "wall_s": 1.614,
    "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": 3,
   "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": 743.6,
    "wall_s": 2.416,
    "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": 3,
   "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": 367.1,
    "wall_s": 1.613,
    "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": 3,
   "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": 306.3,
    "wall_s": 1.934,
    "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": 3,
   "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": 274.8,
    "wall_s": 1.96,
    "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": 3,
   "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": 472.9,
    "wall_s": 2.189,
    "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": 3,
   "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": 568.0,
    "wall_s": 2.056,
    "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": 1,
     "known_claims_blocked": 8,
     "false_blocks": 0,
     "false_passes": 1,
     "unknown_claims": 9,
     "unknown_blocked": 8,
     "non_claims": 11,
     "non_claims_blocked": 0,
     "claim_detection_hits": 18,
     "claim_detection_total": 20,
     "claim_detection_false_alarms": 0,
     "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": 0,
     "claim_detection_hits": 5,
     "claim_detection_total": 6,
     "claim_detection_false_alarms": 0,
     "claim_type_correct": 13
    },
    "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": 1,
     "known_claims_blocked": 6,
     "false_blocks": 0,
     "false_passes": 1,
     "unknown_claims": 2,
     "unknown_blocked": 2,
     "non_claims": 0,
     "non_claims_blocked": 0,
     "claim_detection_hits": 9,
     "claim_detection_total": 10,
     "claim_detection_false_alarms": 0,
     "claim_type_correct": 8
    },
    "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": 0.9885773458255841,
     "probabilities": {
      "none": 0.9978855890274652,
      "completion": 0.001946152825057879,
      "stage": 6.70326190143096e-05,
      "visual": 0.00010122552846250417
     }
    },
    "target_10": {
     "type": "choice",
     "choice": "other",
     "confidence": 0.3399930206205791,
     "probabilities": {
      "deliverable": 0.0058815947804649995,
      "other": 0.5020593260679185,
      "unclear": 0.4920590791516165
     }
    },
    "claim_19": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.9733064173740187,
     "probabilities": {
      "none": 0.9946545713697452,
      "completion": 0.004129073156502436,
      "stage": 0.0006256277974801018,
      "visual": 0.0005907276762722693
     }
    },
    "target_19": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.7039294147401978,
     "probabilities": {
      "deliverable": 0.006056687026034266,
      "other": 0.0841683253592421,
      "unclear": 0.9097749876147236
     }
    },
    "claim_65": {
     "type": "choice",
     "choice": "completion",
     "confidence": 0.8440084651104844,
     "probabilities": {
      "none": 0.0032661707731457307,
      "completion": 0.9497524280116235,
      "stage": 0.0010215819452846478,
      "visual": 0.045959819269946134
     }
    },
    "target_65": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.777342941747574,
     "probabilities": {
      "deliverable": 0.941202685190645,
      "other": 0.05204200149941287,
      "unclear": 0.006755313309942122
     }
    }
   },
   "usage": {
    "input_tokens": 1110,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 542.9,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 0.543
  },
  {
   "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": 0.9958927365730811,
     "probabilities": {
      "none": 0.999331495573339,
      "completion": 0.0006394119303270128,
      "stage": 1.884971001567096e-05,
      "visual": 1.0242786318240456e-05
     }
    },
    "target_1": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.7603941090250831,
     "probabilities": {
      "deliverable": 0.007039735891333927,
      "other": 0.058116008831989224,
      "unclear": 0.9348442552766768
     }
    },
    "claim_10": {
     "type": "choice",
     "choice": "stage",
     "confidence": 0.8307703692134101,
     "probabilities": {
      "none": 0.047650242811773866,
      "completion": 0.006341600343694531,
      "stage": 0.9454158074384311,
      "visual": 0.0005923494061005204
     }
    },
    "target_10": {
     "type": "choice",
     "choice": "other",
     "confidence": 0.45185082871076965,
     "probabilities": {
      "deliverable": 0.03315862121266384,
      "other": 0.792225686143064,
      "unclear": 0.17461569264427218
     }
    },
    "claim_15": {
     "type": "choice",
     "choice": "stage",
     "confidence": 0.5843062615854124,
     "probabilities": {
      "none": 0.005550086999583437,
      "completion": 0.09086064674126894,
      "stage": 0.8369243448920415,
      "visual": 0.06666492136710618
     }
    },
    "target_15": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.205337790765994,
     "probabilities": {
      "deliverable": 0.5119900883019729,
      "other": 0.4288936269389597,
      "unclear": 0.059116284759067446
     }
    },
    "claim_20": {
     "type": "choice",
     "choice": "visual",
     "confidence": 0.28275770627358343,
     "probabilities": {
      "none": 0.1932376061454929,
      "completion": 0.1630155776764689,
      "stage": 0.02245848442551789,
      "visual": 0.6212883317525203
     }
    },
    "target_20": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.014080194142499503,
     "probabilities": {
      "deliverable": 0.4117223025156983,
      "other": 0.269307986935148,
      "unclear": 0.3189697105491537
     }
    },
    "claim_25": {
     "type": "choice",
     "choice": "completion",
     "confidence": 0.5808078603780892,
     "probabilities": {
      "none": 0.23772737259676674,
      "completion": 0.7573955895137453,
      "stage": 0.003135457166282575,
      "visual": 0.0017415807232054195
     }
    },
    "target_25": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.15903101560834088,
     "probabilities": {
      "deliverable": 0.5840286275801185,
      "other": 0.11468061960454723,
      "unclear": 0.30129075281533424
     }
    }
   },
   "usage": {
    "input_tokens": 1856,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 652.1,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.099
  },
  {
   "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": 0.5785629907254665,
     "probabilities": {
      "none": 0.815934465175978,
      "completion": 0.14216089734452014,
      "stage": 0.03991867387825053,
      "visual": 0.0019859636012513086
     }
    },
    "target_2": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.3879207992412338,
     "probabilities": {
      "deliverable": 0.021354198111905053,
      "other": 0.26260279079579263,
      "unclear": 0.7160430110923023
     }
    },
    "claim_11": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.29052526568490356,
     "probabilities": {
      "none": 0.5866585286362468,
      "completion": 0.21563779042681724,
      "stage": 0.19375238895860608,
      "visual": 0.003951291978329805
     }
    },
    "target_11": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.4892042774351619,
     "probabilities": {
      "deliverable": 0.03711759364721769,
      "other": 0.14183919815821558,
      "unclear": 0.8210432081945667
     }
    },
    "claim_24": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.4018651779155824,
     "probabilities": {
      "none": 0.7051289067024892,
      "completion": 0.1637852501417476,
      "stage": 0.12641459915968642,
      "visual": 0.0046712439960767795
     }
    },
    "target_24": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.05606176702648269,
     "probabilities": {
      "deliverable": 0.18855492430824927,
      "other": 0.3435998273138533,
      "unclear": 0.4678452483778974
     }
    },
    "claim_37": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.24707281164195827,
     "probabilities": {
      "none": 0.4256210566627684,
      "completion": 0.16796586614877673,
      "stage": 0.40402813894479167,
      "visual": 0.0023849382436632242
     }
    },
    "target_37": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.1662393594638535,
     "probabilities": {
      "deliverable": 0.08934050891083718,
      "other": 0.3695666232836109,
      "unclear": 0.541092867805552
     }
    },
    "claim_57": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.22580558785674631,
     "probabilities": {
      "none": 0.4896761040822224,
      "completion": 0.24262194604614779,
      "stage": 0.2620974014425804,
      "visual": 0.005604548429049381
     }
    },
    "target_57": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.36530879146491446,
     "probabilities": {
      "deliverable": 0.05411290561365782,
      "other": 0.19915224615923932,
      "unclear": 0.7467348482271029
     }
    },
    "claim_68": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.41046468578703577,
     "probabilities": {
      "none": 0.609436829456601,
      "completion": 0.3446522360239419,
      "stage": 0.04432972989450422,
      "visual": 0.0015812046249527958
     }
    },
    "target_68": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.34659841024393434,
     "probabilities": {
      "deliverable": 0.07086155673711936,
      "other": 0.18369636804546596,
      "unclear": 0.7454420752174147
     }
    }
   },
   "usage": {
    "input_tokens": 2039,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 674.5,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.013
  },
  {
   "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": 0.9969997284277884,
     "probabilities": {
      "none": 0.9995304526793675,
      "completion": 0.0004505995153163534,
      "stage": 1.2788602684586104e-05,
      "visual": 6.159202631494924e-06
     }
    },
    "target_1": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.9748656859210622,
     "probabilities": {
      "deliverable": 0.0010726652520043445,
      "other": 0.0027922499925207194,
      "unclear": 0.996135084755475
     }
    }
   },
   "usage": {
    "input_tokens": 359,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 316.5,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.637
  },
  {
   "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": 0.9956063320748678,
     "probabilities": {
      "none": 0.0007202027896698717,
      "completion": 0.9992675203070353,
      "stage": 6.138451647425906e-06,
      "visual": 6.138451647425906e-06
     }
    },
    "target_3": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.995454561523057,
     "probabilities": {
      "deliverable": 0.9994151556125966,
      "other": 0.0005736620772044575,
      "unclear": 1.118231019878042e-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": 378.4,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.057
  },
  {
   "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": 0.9783796683038555,
     "probabilities": {
      "none": 0.004295962353777458,
      "completion": 0.9954772897093139,
      "stage": 0.00015227733023944986,
      "visual": 7.447060666918633e-05
     }
    },
    "target_5": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9518420705200329,
     "probabilities": {
      "deliverable": 0.9917943996250562,
      "other": 0.005315125297899443,
      "unclear": 0.002890475077044383
     }
    }
   },
   "usage": {
    "input_tokens": 357,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 885.5,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.501
  },
  {
   "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": 0.7186406834479557,
     "probabilities": {
      "none": 0.005530810544680653,
      "completion": 0.11238692922056676,
      "stage": 0.0005992027567260696,
      "visual": 0.8814830574780266
     }
    },
    "target_7": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.7882938066879339,
     "probabilities": {
      "deliverable": 0.939586634254826,
      "other": 0.05956066746526874,
      "unclear": 0.000852698279905309
     }
    }
   },
   "usage": {
    "input_tokens": 361,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 432.0,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.541
  },
  {
   "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": 0.7889388268752012,
     "probabilities": {
      "none": 0.005250349641332756,
      "completion": 0.07051399742792983,
      "stage": 0.0006286629234188279,
      "visual": 0.9236069900073185
     }
    },
    "target_7": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.8001090192748456,
     "probabilities": {
      "deliverable": 0.9513862465399359,
      "other": 0.03824961382721711,
      "unclear": 0.010364139632846978
     }
    }
   },
   "usage": {
    "input_tokens": 359,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 562.1,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.125
  },
  {
   "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": 0.9023554495324727,
     "probabilities": {
      "none": 0.005514417487394744,
      "completion": 0.9749506860411981,
      "stage": 0.0014136203215926297,
      "visual": 0.0181212761498146
     }
    },
    "target_7": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9522400044773742,
     "probabilities": {
      "deliverable": 0.9912200578107884,
      "other": 0.008193692100275207,
      "unclear": 0.0005862500889363427
     }
    }
   },
   "usage": {
    "input_tokens": 360,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 445.5,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.877
  },
  {
   "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": 0.7247997374286497,
     "probabilities": {
      "none": 0.007100693506111981,
      "completion": 0.10332231570414786,
      "stage": 0.0009951458584152428,
      "visual": 0.888581844931325
     }
    },
    "target_7": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.7183477321810019,
     "probabilities": {
      "deliverable": 0.926340588880577,
      "other": 0.023905078344997537,
      "unclear": 0.04975433277442544
     }
    }
   },
   "usage": {
    "input_tokens": 355,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 409.4,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.96
  },
  {
   "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": "none",
     "confidence": 0.12820661450197623,
     "probabilities": {
      "none": 0.44269521938715034,
      "completion": 0.2802803815475616,
      "stage": 0.22368448059458435,
      "visual": 0.05333991847070379
     }
    },
    "target_3": {
     "type": "choice",
     "choice": "other",
     "confidence": 0.20801787180509201,
     "probabilities": {
      "deliverable": 0.4572288187982952,
      "other": 0.48658780117245676,
      "unclear": 0.056183380029248026
     }
    }
   },
   "usage": {
    "input_tokens": 368,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 454.0,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.038
  },
  {
   "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": 0.6766563452812135,
     "probabilities": {
      "none": 0.003518810679681579,
      "completion": 0.1488418971012457,
      "stage": 0.0005695438759978741,
      "visual": 0.8470697483430749
     }
    },
    "target_5": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.7786974038314411,
     "probabilities": {
      "deliverable": 0.9463416277301375,
      "other": 0.036025056286853915,
      "unclear": 0.017633315983008664
     }
    }
   },
   "usage": {
    "input_tokens": 355,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 457.8,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.997
  },
  {
   "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": 0.5220220798795102,
     "probabilities": {
      "none": 0.032895980409747456,
      "completion": 0.21083090481865013,
      "stage": 0.0014996485112027745,
      "visual": 0.7547734662603997
     }
    },
    "target_5": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.7533383068307384,
     "probabilities": {
      "deliverable": 0.9379282831969653,
      "other": 0.04292076786367628,
      "unclear": 0.019150948939358325
     }
    }
   },
   "usage": {
    "input_tokens": 359,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 645.0,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.18
  },
  {
   "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": 0.9950689685719663,
     "probabilities": {
      "none": 0.0008126019075345389,
      "completion": 0.999168078187976,
      "stage": 1.2231653390443583e-05,
      "visual": 7.0882510988369285e-06
     }
    },
    "target_1": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9957684548146749,
     "probabilities": {
      "deliverable": 0.9994769776583994,
      "other": 0.00046887470786542473,
      "unclear": 5.414763373530238e-05
     }
    }
   },
   "usage": {
    "input_tokens": 365,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 317.5,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.614
  },
  {
   "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": 0.9915756874788589,
     "probabilities": {
      "none": 0.9984991855906462,
      "completion": 0.0014003515998033986,
      "stage": 7.88556457219938e-05,
      "visual": 2.1607163828402902e-05
     }
    },
    "target_2": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.11502404719178094,
     "probabilities": {
      "deliverable": 0.1465180396762734,
      "other": 0.3037338954048244,
      "unclear": 0.5497480649189022
     }
    }
   },
   "usage": {
    "input_tokens": 362,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 743.6,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.416
  },
  {
   "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": 0.9426911257893512,
     "probabilities": {
      "none": 0.9868485459195315,
      "completion": 0.008974001448182256,
      "stage": 0.0038286834616105226,
      "visual": 0.0003487691706756358
     }
    },
    "target_2": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.17514012341884233,
     "probabilities": {
      "deliverable": 0.5148135589541735,
      "other": 0.07773306198112048,
      "unclear": 0.407453379064706
     }
    }
   },
   "usage": {
    "input_tokens": 358,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 367.1,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.613
  },
  {
   "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": 0.9681675308909007,
     "probabilities": {
      "none": 0.003442610429788366,
      "completion": 0.9936957989522939,
      "stage": 0.0006734928872529754,
      "visual": 0.0021880977306647986
     }
    },
    "target_3": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9598535461854749,
     "probabilities": {
      "deliverable": 0.9932359075738051,
      "other": 0.005271320633923351,
      "unclear": 0.0014927717922716367
     }
    }
   },
   "usage": {
    "input_tokens": 353,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 306.3,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.934
  },
  {
   "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": 0.9414974642127867,
     "probabilities": {
      "none": 0.005744041735907039,
      "completion": 0.9866576876559577,
      "stage": 0.000260184163550607,
      "visual": 0.007338086444584555
     }
    },
    "target_6": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9757919646526111,
     "probabilities": {
      "deliverable": 0.9962163323352532,
      "other": 0.003145427146076938,
      "unclear": 0.0006382405186699631
     }
    }
   },
   "usage": {
    "input_tokens": 353,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 274.8,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.96
  },
  {
   "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": 0.9435312078926184,
     "probabilities": {
      "none": 0.006909599662493294,
      "completion": 0.9871688361731011,
      "stage": 0.00017454740515841297,
      "visual": 0.00574701675924721
     }
    },
    "target_6": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9712863759570906,
     "probabilities": {
      "deliverable": 0.9953726337917527,
      "other": 0.0038758686682091948,
      "unclear": 0.0007514975400381404
     }
    }
   },
   "usage": {
    "input_tokens": 353,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 472.9,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.189
  },
  {
   "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": 0.9300788624440165,
     "probabilities": {
      "none": 0.007834864479708986,
      "completion": 0.9835525005603705,
      "stage": 0.0004199664943800996,
      "visual": 0.008192668465540471
     }
    },
    "target_5": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9212635286562838,
     "probabilities": {
      "deliverable": 0.9853602960680423,
      "other": 0.00696712153730281,
      "unclear": 0.007672582394654961
     }
    }
   },
   "usage": {
    "input_tokens": 353,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 568.0,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.056
  }
 ],
 "metrics": {
  "requests": 20,
  "completed_requests": 20,
  "input_tokens": 11094,
  "output_tokens": 0,
  "median_latency_ms": 455.9,
  "wall_s": 38.35,
  "cost_usd": 0.0,
  "cost_basis": "free route; missing provider cost fields are not receipts",
  "reported_cost_usd": null
 },
 "wall_s": 38.589
}