{
 "schema_version": 2,
 "project": "evidence-gate",
 "subset": "full",
 "repeat": 2,
 "timestamp": "2026-09-24T16:28:27Z",
 "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": 2,
   "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": 580.0,
    "wall_s": 1.036,
    "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": 2,
   "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": 852.3,
    "wall_s": 2.257,
    "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": 2,
   "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": 708.9,
    "wall_s": 1.853,
    "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": 2,
   "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": 992.4,
    "wall_s": 2.278,
    "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": 2,
   "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": 306.4,
    "wall_s": 1.3,
    "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": 2,
   "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": 364.1,
    "wall_s": 2.053,
    "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": 2,
   "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": 484.1,
    "wall_s": 2.111,
    "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": 2,
   "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": 341.5,
    "wall_s": 1.852,
    "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": 2,
   "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": 339.4,
    "wall_s": 1.991,
    "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": 2,
   "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": 344.4,
    "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": "s07_hedged",
   "group": "synthetic_v1",
   "repeat": 2,
   "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": 434.4,
    "wall_s": 2.083,
    "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": 2,
   "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": 342.7,
    "wall_s": 1.901,
    "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": 2,
   "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": 378.1,
    "wall_s": 2.027,
    "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": 2,
   "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": 370.4,
    "wall_s": 1.984,
    "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": 2,
   "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": 727.5,
    "wall_s": 2.349,
    "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": 2,
   "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": 358,
    "output_tokens": 0,
    "median_latency_ms": 323.3,
    "wall_s": 1.588,
    "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": 2,
   "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": 275.7,
    "wall_s": 1.944,
    "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": 2,
   "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": 321.7,
    "wall_s": 2.033,
    "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": 2,
   "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": 326.9,
    "wall_s": 1.993,
    "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": 2,
   "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": 353.6,
    "wall_s": 2.01,
    "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": 0.9890413985166999,
     "probabilities": {
      "none": 0.9979913654810703,
      "completion": 0.0018323683603295555,
      "stage": 0.0001150584911311209,
      "visual": 6.120766746902188e-05
     }
    },
    "target_10": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.3564981047755358,
     "probabilities": {
      "deliverable": 0.004830360553123989,
      "other": 0.4160581717256043,
      "unclear": 0.5791114677212718
     }
    },
    "claim_19": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.9727487440994773,
     "probabilities": {
      "none": 0.994481546032509,
      "completion": 0.004385272842649003,
      "stage": 0.0006263262902631995,
      "visual": 0.0005068548345787706
     }
    },
    "target_19": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.7118883295434488,
     "probabilities": {
      "deliverable": 0.005431228160554198,
      "other": 0.08184170763417546,
      "unclear": 0.9127270642052703
     }
    },
    "claim_65": {
     "type": "choice",
     "choice": "completion",
     "confidence": 0.8539604737750583,
     "probabilities": {
      "none": 0.003349175010182777,
      "completion": 0.9550427653877185,
      "stage": 0.0017137963443478752,
      "visual": 0.03989426325775079
     }
    },
    "target_65": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.7909628597080627,
     "probabilities": {
      "deliverable": 0.9451675639663462,
      "other": 0.04965418427078058,
      "unclear": 0.00517825176287325
     }
    }
   },
   "usage": {
    "input_tokens": 1110,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 580.0,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.036
  },
  {
   "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": 0.9953640077746913,
     "probabilities": {
      "none": 0.9992280371898149,
      "completion": 0.0007478548789197468,
      "stage": 1.574583387921662e-05,
      "visual": 8.36209738612708e-06
     }
    },
    "target_1": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.8310548928821966,
     "probabilities": {
      "deliverable": 0.00553518070815641,
      "other": 0.03494091827321568,
      "unclear": 0.9595239010186278
     }
    },
    "claim_10": {
     "type": "choice",
     "choice": "stage",
     "confidence": 0.767890077353817,
     "probabilities": {
      "none": 0.08471651119689942,
      "completion": 0.00468978730978985,
      "stage": 0.9103532636153603,
      "visual": 0.0002404378779504333
     }
    },
    "target_10": {
     "type": "choice",
     "choice": "other",
     "confidence": 0.48245250027460607,
     "probabilities": {
      "deliverable": 0.043200790017714444,
      "other": 0.8211164976393088,
      "unclear": 0.13568271234297669
     }
    },
    "claim_15": {
     "type": "choice",
     "choice": "stage",
     "confidence": 0.591549201109996,
     "probabilities": {
      "none": 0.005308258738960466,
      "completion": 0.1707446610888913,
      "stage": 0.8084155942451856,
      "visual": 0.015531485926962697
     }
    },
    "target_15": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.27287435847462893,
     "probabilities": {
      "deliverable": 0.6550904704722477,
      "other": 0.2879275172016107,
      "unclear": 0.05698201232614163
     }
    },
    "claim_20": {
     "type": "choice",
     "choice": "visual",
     "confidence": 0.2036070003417474,
     "probabilities": {
      "none": 0.40892013455386617,
      "completion": 0.1522597820711491,
      "stage": 0.023048621392540006,
      "visual": 0.41577146198244475
     }
    },
    "target_20": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.08964585825891103,
     "probabilities": {
      "deliverable": 0.5402166700837523,
      "other": 0.2739088267609776,
      "unclear": 0.18587450315527013
     }
    },
    "claim_25": {
     "type": "choice",
     "choice": "completion",
     "confidence": 0.6849992470818709,
     "probabilities": {
      "none": 0.13863017583287562,
      "completion": 0.856270680263408,
      "stage": 0.0037215982557284705,
      "visual": 0.0013775456479879335
     }
    },
    "target_25": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.2774251784729772,
     "probabilities": {
      "deliverable": 0.6846197616461803,
      "other": 0.07288131432013617,
      "unclear": 0.24249892403368353
     }
    }
   },
   "usage": {
    "input_tokens": 1856,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 852.3,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.257
  },
  {
   "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": 0.5384053273792484,
     "probabilities": {
      "none": 0.7848778822709409,
      "completion": 0.17154741858276162,
      "stage": 0.0406077782099322,
      "visual": 0.002966920936365219
     }
    },
    "target_2": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.41438400350165894,
     "probabilities": {
      "deliverable": 0.025866404380590798,
      "other": 0.22237905055526622,
      "unclear": 0.751754545064143
     }
    },
    "claim_11": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.2443650727425265,
     "probabilities": {
      "none": 0.5190556588593955,
      "completion": 0.22281503869605804,
      "stage": 0.2535394492734997,
      "visual": 0.004589853171046775
     }
    },
    "target_11": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.45560367598680096,
     "probabilities": {
      "deliverable": 0.046691044732276615,
      "other": 0.14675006137281557,
      "unclear": 0.8065588938949078
     }
    },
    "claim_24": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.32739506842006527,
     "probabilities": {
      "none": 0.6305691029944567,
      "completion": 0.17949763507408845,
      "stage": 0.18633036648686338,
      "visual": 0.0036028954445915294
     }
    },
    "target_24": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.10019337140221585,
     "probabilities": {
      "deliverable": 0.1896350093182929,
      "other": 0.25265520231942684,
      "unclear": 0.5577097883622801
     }
    },
    "claim_37": {
     "type": "choice",
     "choice": "stage",
     "confidence": 0.23932500858437933,
     "probabilities": {
      "none": 0.35240179608988653,
      "completion": 0.18642992900524513,
      "stage": 0.45840741637737165,
      "visual": 0.0027608585274966875
     }
    },
    "target_37": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.2377127691462112,
     "probabilities": {
      "deliverable": 0.07774988666450845,
      "other": 0.2776805077026838,
      "unclear": 0.6445696056328077
     }
    },
    "claim_57": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.21993085851632277,
     "probabilities": {
      "none": 0.4778563668411234,
      "completion": 0.25646925788176433,
      "stage": 0.26001064438513344,
      "visual": 0.005663730891978896
     }
    },
    "target_57": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.42730477377050247,
     "probabilities": {
      "deliverable": 0.04994452900074151,
      "other": 0.16006441779609307,
      "unclear": 0.7899910532031654
     }
    },
    "claim_68": {
     "type": "choice",
     "choice": "none",
     "confidence": 0.3828067093518398,
     "probabilities": {
      "none": 0.4813414403062143,
      "completion": 0.4732578891637619,
      "stage": 0.043058261258795076,
      "visual": 0.0023424092712287393
     }
    },
    "target_68": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.2572374834552158,
     "probabilities": {
      "deliverable": 0.16538718977266384,
      "other": 0.13387476745681698,
      "unclear": 0.7007380427705192
     }
    }
   },
   "usage": {
    "input_tokens": 2039,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 708.9,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.853
  },
  {
   "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": 0.9981258073425303,
     "probabilities": {
      "none": 0.9997228001113152,
      "completion": 0.00026695590639939427,
      "stage": 6.540116672769304e-06,
      "visual": 3.7038656126752254e-06
     }
    },
    "target_1": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.9696871997766946,
     "probabilities": {
      "deliverable": 0.0013656654439700199,
      "other": 0.003437737148256886,
      "unclear": 0.995196597407773
     }
    }
   },
   "usage": {
    "input_tokens": 359,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 992.4,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.278
  },
  {
   "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": 0.9922798212144651,
     "probabilities": {
      "none": 0.0013615853278697828,
      "completion": 0.9986090021702386,
      "stage": 1.3407411077920278e-05,
      "visual": 1.6005090813786182e-05
     }
    },
    "target_3": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9892295490265787,
     "probabilities": {
      "deliverable": 0.9984258509758325,
      "other": 0.0015557798766315581,
      "unclear": 1.836914753602911e-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": 306.4,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.3
  },
  {
   "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": 0.9831874208503978,
     "probabilities": {
      "none": 0.003248750590569997,
      "completion": 0.9966139855705768,
      "stage": 8.213306798348238e-05,
      "visual": 5.5130770869709424e-05
     }
    },
    "target_5": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9527108906680154,
     "probabilities": {
      "deliverable": 0.9919819892424548,
      "other": 0.005074264444202395,
      "unclear": 0.0029437463133426835
     }
    }
   },
   "usage": {
    "input_tokens": 357,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 364.1,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.053
  },
  {
   "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": 0.7228165871663659,
     "probabilities": {
      "none": 0.00730362576084009,
      "completion": 0.10499660414295743,
      "stage": 0.0007334946037261666,
      "visual": 0.8869662754924763
     }
    },
    "target_7": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.750936244975363,
     "probabilities": {
      "deliverable": 0.9239638489237588,
      "other": 0.07519067355336606,
      "unclear": 0.0008454775228751712
     }
    }
   },
   "usage": {
    "input_tokens": 361,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 484.1,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.111
  },
  {
   "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": 0.7447099991408735,
     "probabilities": {
      "none": 0.005216785620276618,
      "completion": 0.09609820851735254,
      "stage": 0.0006554005215927822,
      "visual": 0.8980296053407781
     }
    },
    "target_7": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.781299504751084,
     "probabilities": {
      "deliverable": 0.9458245008515292,
      "other": 0.04136167710487715,
      "unclear": 0.012813822043593666
     }
    }
   },
   "usage": {
    "input_tokens": 359,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 341.5,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.852
  },
  {
   "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": 0.9469357371603975,
     "probabilities": {
      "none": 0.005631550111457278,
      "completion": 0.9883764008307745,
      "stage": 0.0007076272338447171,
      "visual": 0.005284421823923562
     }
    },
    "target_7": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9598154270830482,
     "probabilities": {
      "deliverable": 0.9928762907422505,
      "other": 0.00661721717103123,
      "unclear": 0.0005064920867182387
     }
    }
   },
   "usage": {
    "input_tokens": 360,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 339.4,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.991
  },
  {
   "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": 0.5920792073092649,
     "probabilities": {
      "none": 0.013951660495263503,
      "completion": 0.18871231917852604,
      "stage": 0.0014552607466992954,
      "visual": 0.7958807595795111
     }
    },
    "target_7": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.6751151558335146,
     "probabilities": {
      "deliverable": 0.9107604639614131,
      "other": 0.028832668003737724,
      "unclear": 0.06040686803484923
     }
    }
   },
   "usage": {
    "input_tokens": 355,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 344.4,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.997
  },
  {
   "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": 0.2248507965817832,
     "probabilities": {
      "none": 0.2943600722535576,
      "completion": 0.5247976338255873,
      "stage": 0.1588297101364112,
      "visual": 0.022012583784443963
     }
    },
    "target_3": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.2942466241074877,
     "probabilities": {
      "deliverable": 0.6573286535650882,
      "other": 0.29804291374236525,
      "unclear": 0.044628432692546564
     }
    }
   },
   "usage": {
    "input_tokens": 368,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 434.4,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.083
  },
  {
   "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": 0.7083570692387626,
     "probabilities": {
      "none": 0.004220334333621723,
      "completion": 0.12380220680016639,
      "stage": 0.00035698208937721856,
      "visual": 0.8716204767768347
     }
    },
    "target_5": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.7944836374420161,
     "probabilities": {
      "deliverable": 0.9481852385822208,
      "other": 0.04401032743517365,
      "unclear": 0.00780443398260552
     }
    }
   },
   "usage": {
    "input_tokens": 355,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 342.7,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.901
  },
  {
   "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": 0.6309545977751306,
     "probabilities": {
      "none": 0.03420295591563137,
      "completion": 0.1168992645124872,
      "stage": 0.0007715211712256448,
      "visual": 0.8481262584006558
     }
    },
    "target_5": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.7809746844876455,
     "probabilities": {
      "deliverable": 0.9450081691117166,
      "other": 0.04388704081683352,
      "unclear": 0.011104790071449934
     }
    }
   },
   "usage": {
    "input_tokens": 359,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 378.1,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.027
  },
  {
   "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": 0.9953673136049256,
     "probabilities": {
      "none": 0.0007653078310832776,
      "completion": 0.9992217963065428,
      "stage": 8.982257762425716e-06,
      "visual": 3.913604611573418e-06
     }
    },
    "target_1": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9957515908327004,
     "probabilities": {
      "deliverable": 0.9994728382383461,
      "other": 0.0004789245178611327,
      "unclear": 4.8237243792739745e-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": 370.4,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.984
  },
  {
   "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": 0.9957688240337784,
     "probabilities": {
      "none": 0.9993094950399286,
      "completion": 0.0006587863408394336,
      "stage": 1.949464529120118e-05,
      "visual": 1.2223973940705684e-05
     }
    },
    "target_2": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.13616481636248945,
     "probabilities": {
      "deliverable": 0.13263041164647818,
      "other": 0.2974156098531282,
      "unclear": 0.5699539785003936
     }
    }
   },
   "usage": {
    "input_tokens": 362,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 727.5,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.349
  },
  {
   "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": 0.9460579199416014,
     "probabilities": {
      "none": 0.9875365774580567,
      "completion": 0.009332613954428212,
      "stage": 0.0029404069973468783,
      "visual": 0.0001904015901682761
     }
    },
    "target_2": {
     "type": "choice",
     "choice": "unclear",
     "confidence": 0.15923990323961967,
     "probabilities": {
      "deliverable": 0.43750230010334895,
      "other": 0.08471045812977857,
      "unclear": 0.4777872417668725
     }
    }
   },
   "usage": {
    "input_tokens": 358,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 323.3,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.588
  },
  {
   "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": 0.9451393185399662,
     "probabilities": {
      "none": 0.0034630132154459536,
      "completion": 0.9877229693600883,
      "stage": 0.0007347798086422839,
      "visual": 0.008079237615823569
     }
    },
    "target_3": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9561906077964019,
     "probabilities": {
      "deliverable": 0.9925314008113472,
      "other": 0.005676990003330225,
      "unclear": 0.0017916091853226392
     }
    }
   },
   "usage": {
    "input_tokens": 353,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 275.7,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.944
  },
  {
   "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": 0.9455210418950777,
     "probabilities": {
      "none": 0.0056833012604741925,
      "completion": 0.9877127269743909,
      "stage": 0.0001661706399433375,
      "visual": 0.006437801125191469
     }
    },
    "target_6": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9730073439050898,
     "probabilities": {
      "deliverable": 0.9957397944622491,
      "other": 0.00339548398474808,
      "unclear": 0.00086472155300285
     }
    }
   },
   "usage": {
    "input_tokens": 353,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 321.7,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.033
  },
  {
   "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": 0.9384967592943493,
     "probabilities": {
      "none": 0.006183019977666746,
      "completion": 0.985768876094632,
      "stage": 0.00018349293052291116,
      "visual": 0.007864610997178237
     }
    },
    "target_6": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9739322060403194,
     "probabilities": {
      "deliverable": 0.9958911631031095,
      "other": 0.003352839436136445,
      "unclear": 0.0007559974607541399
     }
    }
   },
   "usage": {
    "input_tokens": 353,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 326.9,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 1.993
  },
  {
   "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": 0.9232858892195956,
     "probabilities": {
      "none": 0.01089745518284184,
      "completion": 0.981447951328874,
      "stage": 0.0003510221116608647,
      "visual": 0.007303571376623301
     }
    },
    "target_5": {
     "type": "choice",
     "choice": "deliverable",
     "confidence": 0.9099924404278242,
     "probabilities": {
      "deliverable": 0.9826627697254268,
      "other": 0.010974139698192142,
      "unclear": 0.006363090576381139
     }
    }
   },
   "usage": {
    "input_tokens": 353,
    "output_tokens": 0
   },
   "model": "openjev-0.1",
   "mode": "live",
   "provider": "codiv",
   "label": "openjev (Codiv free hosted) - NOT JEV",
   "latency_ms": 353.6,
   "cached": false,
   "reported_cost_usd": null,
   "status": "ok",
   "elapsed_s": 2.01
  }
 ],
 "metrics": {
  "requests": 20,
  "completed_requests": 20,
  "input_tokens": 11094,
  "output_tokens": 0,
  "median_latency_ms": 358.9,
  "wall_s": 38.64,
  "cost_usd": 0.0,
  "cost_basis": "free route; missing provider cost fields are not receipts",
  "reported_cost_usd": null
 },
 "wall_s": 38.83
}