training_evidence/run_B_forced_eval_track/results.json

{
  "scripted": {
    "policy": "scripted",
    "n_scenarios_tech": 24,
    "n_scenarios_meridian": 12,
    "mean_reward_tech": -0.025,
    "prediction_accuracy": 1.0,
    "catastrophe_count": 0,
    "catastrophe_rate": 0.0
  },
  "sft_only": {
    "policy": "sft_only",
    "n_scenarios_tech": 24,
    "n_scenarios_meridian": 12,
    "mean_reward_tech": 0.4057,
    "prediction_accuracy": 1.0,
    "catastrophe_count": 0,
    "catastrophe_rate": 0.0
  },
  "grpo_trained": {
    "policy": "grpo_trained",
    "n_scenarios_tech": 24,
    "n_scenarios_meridian": 12,
    "mean_reward_tech": 0.6275,
    "prediction_accuracy": 0.7083,
    "catastrophe_count": 0,
    "catastrophe_rate": 0.0
  }
}