artifacts/eval_results.json

{
  "baseline": {
    "n": 50,
    "mean_reward": 0.9,
    "success_rate": 0.5
  },
  "trained": {
    "n": 50,
    "mean_reward": 1.5120000000000002,
    "success_rate": 0.86
  },
  "plots": [
    "baseline_vs_trained.png",
    "training_reward_curve.png",
    "success_by_category.png"
  ],
  "repair_library_size": 43
}