{
  "status": "passed",
  "reference_questions": 40,
  "changed_decisions": 0,
  "max_probability_difference": 2.9895792783563024e-06,
  "probability_tolerance": 0.0001,
  "precision": "float32",
  "versions": {
    "torch": "2.9.1+cpu",
    "transformers": "4.57.6"
  },
  "scope": "40 fixed validation questions: 32 Maze movement and 8 ViZDoom. Runtime conversion check, not a new task benchmark."
}