{
  "status": "completed",
  "source": "completed GPU generation pilot artifacts",
  "arms": {
    "base": {
      "problems": 8,
      "samples_per_problem": 1,
      "records": 8,
      "accuracy_over_samples": 0.0,
      "truncated": 8,
      "no_boxed": 8,
      "generation_wall_s": 642.22137600102,
      "session_wall_s": 671.5753352389438
    },
    "edited": {
      "problems": 8,
      "samples_per_problem": 4,
      "records": 32,
      "accuracy_over_samples": 0.25,
      "cluster_mean_accuracy": 0.25,
      "cluster_bootstrap_95ci": [0.0625, 0.46875],
      "truncated": 18,
      "no_boxed": 20,
      "generation_wall_s": 1177.7240174530307,
      "session_wall_s": 1294.2062180310022
    }
  },
  "comparison_note": "Descriptive only: the arms have different samples per problem (base 1, edited 4), so this is not a matched-sample causal comparison.",
  "gpu_compute_time": null
}
