{
  "protocol": "v0.6.2",
  "task_count": 100,
  "family_count": 5,
  "development_trials": 200,
  "main_trials": 1000,
  "repeat_trials": 100,
  "temperature": 0,
  "source": "Frozen reports/v0.6.2/evaluation_analysis.json and manifests/v0.6.2/evaluation_manifest.json; verified against 10 saved model-condition totals.",
  "models": {
    "mini": {
      "name": "GPT-4.1 mini",
      "snapshot": "gpt-4.1-mini-2025-04-14",
      "both_correct": 100,
      "loss": 0,
      "gain": 0,
      "both_fail": 0,
      "different_trajectories": 13
    },
    "full": {
      "name": "GPT-4.1",
      "snapshot": "gpt-4.1-2025-04-14",
      "both_correct": 98,
      "loss": 1,
      "gain": 1,
      "both_fail": 0,
      "different_trajectories": 6
    }
  },
  "conditions": [
    {
      "id": "no_memory",
      "label": "No memory",
      "description": "No previous trajectory is supplied.",
      "mini": 98,
      "full": 98
    },
    {
      "id": "aligned_similar",
      "label": "Aligned similar",
      "description": "A similar successful source whose demonstrated choice fits the target.",
      "mini": 100,
      "full": 100
    },
    {
      "id": "irrelevant",
      "label": "Ordinary irrelevant",
      "description": "A successful source from another workflow; a secondary control.",
      "mini": 99,
      "full": 98
    },
    {
      "id": "misaligned_similar",
      "label": "Misaligned similar",
      "description": "A similar successful source whose specific choice does not fit the current case.",
      "mini": 100,
      "full": 99
    },
    {
      "id": "irrelevant_action_matched",
      "label": "Action-matched control",
      "description": "A less-similar successful source with matched action count, tool-name multiset, and current-tool overlap.",
      "mini": 100,
      "full": 99
    }
  ],
  "notes": [
    "Critical-decision correctness and final success have the same aggregate count in each arm.",
    "First-action correctness is 100/100 in every model-condition arm.",
    "Repeat trials are excluded from primary denominators.",
    "This website exports selected excerpts and reported aggregates, not the full frozen dataset."
  ]
}