{
  "model_name": "AlphaApple FruitBox AI",
  "description": "PPO agent trained to play Korean fruit box puzzle game",
  "architecture": "CNN + PPO",
  "input_shape": [
    17,
    10,
    1
  ],
  "output_shape": "action_logits + value",
  "training_steps": 1000000,
  "performance": {
    "average_score": 77.0,
    "vs_random": "+7.1%",
    "vs_greedy": "+5.0%"
  }
}