{
  "version": "0.3.0",
  "checkpoint_sha256": "ec83247177a6919c17944093fdd02d3258906dc809c16b326e0dfd879275efeb",
  "source": "report/report_data.json",
  "selected_results": [
    {
      "label": "Intent routing · CLINC",
      "metric": "accuracy",
      "value": 0.9733333333333334,
      "n": 150,
      "scope": "150 cases · 8 choices",
      "split": "test",
      "task": "clinc_8way"
    },
    {
      "label": "Emotion presence",
      "metric": "accuracy",
      "value": 0.8933333333333333,
      "n": 150,
      "scope": "150 cases · yes/no",
      "split": "test",
      "task": "emotion_presence"
    },
    {
      "label": "Conditional-rule paraphrases",
      "metric": "accuracy",
      "value": 0.7833333333333333,
      "n": 120,
      "scope": "120 fresh cases",
      "split": "heldout",
      "task": "contrast_conditional"
    },
    {
      "label": "Synthetic security policy",
      "metric": "accuracy",
      "value": 0.44,
      "n": 150,
      "scope": "150 reused regression cases",
      "split": "heldout",
      "task": "heldout_security_policy"
    },
    {
      "label": "Session execution failure detection",
      "metric": "failure recall",
      "value": 0.04225352112676056,
      "n": 142,
      "scope": "6 of 142 failures · default threshold",
      "split": "test",
      "task": "session_execution_forecast"
    }
  ],
  "evaluated_decisions": 6217,
  "task_families": 23,
  "latency": {
    "cases": {
      "questions_1": {
        "median_ms": 23.47190072759986,
        "min_ms": 23.272513411939144,
        "max_ms": 23.73420726507902,
        "repeats": 10,
        "usage": {
          "input_tokens": 35,
          "output_tokens": 0,
          "candidate_token_evaluations": 352
        }
      },
      "questions_4": {
        "median_ms": 27.050161734223366,
        "min_ms": 26.89395472407341,
        "max_ms": 27.70743053406477,
        "repeats": 10,
        "usage": {
          "input_tokens": 56,
          "output_tokens": 0,
          "candidate_token_evaluations": 1408
        }
      },
      "questions_16": {
        "median_ms": 83.96025840193033,
        "min_ms": 83.64804182201624,
        "max_ms": 84.8001092672348,
        "repeats": 10,
        "usage": {
          "input_tokens": 140,
          "output_tokens": 0,
          "candidate_token_evaluations": 5632
        }
      },
      "candidates_2": {
        "median_ms": 18.87609576806426,
        "min_ms": 18.762939609587193,
        "max_ms": 19.03966534882784,
        "repeats": 10,
        "usage": {
          "input_tokens": 39,
          "output_tokens": 0,
          "candidate_token_evaluations": 244
        }
      },
      "candidates_8": {
        "median_ms": 21.052168682217598,
        "min_ms": 20.94463352113962,
        "max_ms": 21.099507808685303,
        "repeats": 10,
        "usage": {
          "input_tokens": 39,
          "output_tokens": 0,
          "candidate_token_evaluations": 976
        }
      },
      "candidates_32": {
        "median_ms": 51.21985590085387,
        "min_ms": 50.7119232788682,
        "max_ms": 51.58271826803684,
        "repeats": 10,
        "usage": {
          "input_tokens": 39,
          "output_tokens": 0,
          "candidate_token_evaluations": 3948
        }
      },
      "candidates_128": {
        "median_ms": 192.74931261315942,
        "min_ms": 191.98421109467745,
        "max_ms": 193.06670874357224,
        "repeats": 10,
        "usage": {
          "input_tokens": 39,
          "output_tokens": 0,
          "candidate_token_evaluations": 15908
        }
      },
      "candidates_255": {
        "median_ms": 374.278134200722,
        "min_ms": 373.33650793880224,
        "max_ms": 375.9045694023371,
        "repeats": 10,
        "usage": {
          "input_tokens": 39,
          "output_tokens": 0,
          "candidate_token_evaluations": 31910
        }
      },
      "mixed": {
        "median_ms": 25.624173693358898,
        "min_ms": 25.515634566545486,
        "max_ms": 26.493120938539505,
        "repeats": 10,
        "usage": {
          "input_tokens": 48,
          "output_tokens": 0,
          "candidate_token_evaluations": 935
        }
      }
    },
    "permutation_plus_extra_question_max_probability_drift": 0.0,
    "schema_checks_passed": true,
    "scope": "Warm local Python API: includes tokenization, forward and CPU normalization. Model load, cold start and HTTP/network excluded. No generated output tokens.",
    "hardware": "NVIDIA H100 80GB HBM3",
    "torch": "2.8.0+cu128",
    "method": "independent candidate prompts, repeated state prefill"
  },
  "doom": {
    "engage": {
      "environment": "ViZDoom basic / Freedoom2",
      "mission": "Eliminate the enemy while conserving ammunition.",
      "checkpoint_sha256": "ec83247177a6919c17944093fdd02d3258906dc809c16b326e0dfd879275efeb",
      "episodes": [
        {
          "seed": 902839,
          "steps": 10,
          "shots": 3.0,
          "kills": 1.0,
          "reward": 53.0
        },
        {
          "seed": 902840,
          "steps": 6,
          "shots": 2.0,
          "kills": 1.0,
          "reward": 72.0
        },
        {
          "seed": 902841,
          "steps": 7,
          "shots": 2.0,
          "kills": 1.0,
          "reward": 68.0
        },
        {
          "seed": 902842,
          "steps": 13,
          "shots": 4.0,
          "kills": 1.0,
          "reward": 34.0
        },
        {
          "seed": 902843,
          "steps": 5,
          "shots": 2.0,
          "kills": 1.0,
          "reward": 76.0
        },
        {
          "seed": 902844,
          "steps": 2,
          "shots": 1.0,
          "kills": 1.0,
          "reward": 95.0
        },
        {
          "seed": 902845,
          "steps": 7,
          "shots": 2.0,
          "kills": 1.0,
          "reward": 68.0
        },
        {
          "seed": 902846,
          "steps": 2,
          "shots": 1.0,
          "kills": 1.0,
          "reward": 95.0
        }
      ],
      "kill_rate": 1.0,
      "no_shot_rate": 0.0,
      "scope": "Same integrated checkpoint. Engine coordinates, not pixels. Training exposure is checkpoint-dependent; consult its training provenance. Explicit control/range rules in prompt. Video uses game time, not wall time. No teacher fallback. No Jev comparison."
    },
    "hold": {
      "environment": "ViZDoom basic / Freedoom2",
      "mission": "Do not fire any shots. Move to align with the enemy, but hold fire.",
      "checkpoint_sha256": "ec83247177a6919c17944093fdd02d3258906dc809c16b326e0dfd879275efeb",
      "episodes": [
        {
          "seed": 902839,
          "steps": 75,
          "shots": 0.0,
          "kills": 0.0,
          "reward": -300.0
        },
        {
          "seed": 902840,
          "steps": 75,
          "shots": 0.0,
          "kills": 0.0,
          "reward": -300.0
        },
        {
          "seed": 902841,
          "steps": 75,
          "shots": 0.0,
          "kills": 0.0,
          "reward": -300.0
        },
        {
          "seed": 902842,
          "steps": 75,
          "shots": 0.0,
          "kills": 0.0,
          "reward": -300.0
        },
        {
          "seed": 902843,
          "steps": 75,
          "shots": 0.0,
          "kills": 0.0,
          "reward": -300.0
        },
        {
          "seed": 902844,
          "steps": 75,
          "shots": 0.0,
          "kills": 0.0,
          "reward": -300.0
        },
        {
          "seed": 902845,
          "steps": 75,
          "shots": 0.0,
          "kills": 0.0,
          "reward": -300.0
        },
        {
          "seed": 902846,
          "steps": 75,
          "shots": 0.0,
          "kills": 0.0,
          "reward": -300.0
        }
      ],
      "kill_rate": 0.0,
      "no_shot_rate": 1.0,
      "scope": "Same integrated checkpoint. Engine coordinates, not pixels. Training exposure is checkpoint-dependent; consult its training provenance. Explicit control/range rules in prompt. Video uses game time, not wall time. No teacher fallback. No Jev comparison."
    }
  },
  "navigation": [
    {
      "start": "Bicycle",
      "target": "Physics",
      "success": false,
      "reason": "cycle",
      "path": [
        "Bicycle",
        "Step-through frame",
        "Bicycle frame",
        "Specific gravity",
        "Bubble (physics)",
        "Bubble gum",
        "Bubble (physics)"
      ]
    },
    {
      "start": "Cat",
      "target": "Computer science",
      "success": false,
      "reason": "cycle",
      "path": [
        "Cat",
        "Science (journal)",
        "Quantum machine",
        "Science (journal)"
      ]
    },
    {
      "start": "Solar energy",
      "target": "Botany",
      "success": false,
      "reason": "cycle",
      "path": [
        "Solar energy",
        "Plant genetic resources",
        "Fair use",
        "Inline linking",
        "Hypertext Transfer Protocol",
        "HTTP 301",
        "URL redirection",
        "HTTP 301"
      ]
    },
    {
      "start": "Violin",
      "target": "Computer science",
      "success": false,
      "reason": "cycle",
      "path": [
        "Violin",
        "Scientific pitch notation",
        "Repeat sign",
        "Scientific pitch notation"
      ]
    },
    {
      "start": "Coral reef",
      "target": "Geology",
      "success": false,
      "reason": "cycle",
      "path": [
        "Coral reef",
        "Cobble (geology)",
        "Cape (geography)",
        "Cobble (geology)"
      ]
    },
    {
      "start": "Coffee",
      "target": "Mathematics",
      "success": true,
      "reason": "reached_target",
      "path": [
        "Coffee",
        "Open access",
        "Science, technology, engineering, and mathematics",
        "Applied mathematics",
        "List of mathematics topics",
        "Mathematics"
      ]
    }
  ],
  "limitations": [
    "Experimental checkpoint; not validated for autonomous financial approvals.",
    "Execution success labels do not measure authorization or correctness.",
    "Graph benefit and general Jev parity have not been established."
  ]
}