{
  "method": "DEBUG TRIAL \u2014 REINFORCE v3 (fresh LoRA, heuristic override)",
  "base_model": "Qwen/Qwen3-4B",
  "config": {
    "episodes": 3,
    "max_steps": 15,
    "lr": 5e-05,
    "temperature": 1.0
  },
  "gpu": "NVIDIA L40S",
  "vram_gb": 47.7,
  "episodes": [
    {
      "episode": 0,
      "task": "easy_bench",
      "total_reward": 6.6147,
      "enrolled": 5,
      "target": 80,
      "steps": 15,
      "unique_action_types": 2,
      "positive_rl_steps": 15,
      "action_dist": {
        "screen_patient": 6,
        "allocate_to_site": 9
      }
    },
    {
      "episode": 1,
      "task": "easy_bench",
      "total_reward": 6.6147,
      "enrolled": 5,
      "target": 80,
      "steps": 15,
      "unique_action_types": 2,
      "positive_rl_steps": 15,
      "action_dist": {
        "screen_patient": 6,
        "allocate_to_site": 9
      }
    },
    {
      "episode": 2,
      "task": "easy_bench",
      "total_reward": 7.4131,
      "enrolled": 1,
      "target": 80,
      "steps": 15,
      "unique_action_types": 2,
      "positive_rl_steps": 15,
      "action_dist": {
        "screen_patient": 14,
        "allocate_to_site": 1
      }
    }
  ]
}