{
  "perception_radius": 2.5,
  "local_knowledge": true,
  "training_seed": 712,
  "heldout_seeds": [
    9000,
    9001,
    9002,
    9003,
    9004,
    9005,
    9006,
    9007,
    9008,
    9009,
    9010,
    9011,
    9012,
    9013,
    9014,
    9015,
    9016,
    9017,
    9018,
    9019
  ],
  "architecture": "RecurrentPPO LSTM 32 + MLP 32, shared weights, per-bee states",
  "bootstrap": "1800 gradient steps imitation of a simple heuristic, followed by 65536 PPO steps",
  "scores": {
    "neural": {
      "mean_delivered_g": 0.6493469108141763,
      "per_seed": [
        1.1273730892166423,
        0.64,
        0.6653188336148164,
        0.39999999999999997,
        0.36,
        0.68,
        0.8400000000000002,
        0.43999999999999995,
        1.1600000000000004,
        0.660944871983306,
        0.4799999999999999,
        0.9149082845092122,
        0.4799999999999999,
        0.4799999999999999,
        0.3758822082115054,
        0.5481025676673433,
        0.7600000000000001,
        0.7600000000000001,
        0.7451725514974283,
        0.46923580958326894
      ]
    },
    "heuristic": {
      "mean_delivered_g": 0.6559086275748716,
      "per_seed": [
        1.1600000000000004,
        0.64,
        0.7200000000000001,
        0.39999999999999997,
        0.36,
        0.6,
        0.8000000000000002,
        0.43999999999999995,
        1.2400000000000004,
        0.68,
        0.4799999999999999,
        0.9600000000000003,
        0.4799999999999999,
        0.43999999999999995,
        0.32,
        0.5599999999999999,
        0.8000000000000002,
        0.8000000000000002,
        0.7581725514974285,
        0.4799999999999999
      ]
    },
    "random": {
      "mean_delivered_g": 0.031000000000000007,
      "per_seed": [
        0.0,
        0.0,
        0.08,
        0.0,
        0.0,
        0.04,
        0.04,
        0.04,
        0.12000000000000001,
        0.0,
        0.04,
        0.02,
        0.0,
        0.04,
        0.04,
        0.04,
        0.0,
        0.04,
        0.04,
        0.04
      ]
    }
  },
  "passed": true,
  "limitations": "Experimental environment, not biological validation. Frozen weights at runtime.",
  "policy_hash": "278c39db4138ffef"
}