artifacts/evals/eval_protocol.json

{
  "task_set": "easy_001..hard_015",
  "seed": 42,
  "max_steps": 12,
  "decode_config": {
    "temperature": 0.0,
    "top_p": 1.0,
    "max_new_tokens": 256
  },
  "action_parser": "CommitmentAction pydantic schema"
}