{
  "judge_model": "gpt-5.4",
  "reasoning_effort": "none",
  "part1_policy_prompt_limit": 128,
  "part2_policy_prompt_limit": 128,
  "part1": {
    "reward_model_pair_accuracy": 0.74,
    "dpo": 0.65,
    "ipo": 0.60,
    "aot": 0.65,
    "grpo": 0.60,
    "drgrpo": 0.60,
    "gspo": 0.60
  },
  "part2": {
    "offline_policy_win_rate": 0.82,
    "online_policy_win_rate": 0.70
  }
}