{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.006,
      "degree": 2
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "degree": 2
        },
        "mean": 0.003609913313994184
      },
      {
        "cfg": {
          "lr": 0.003,
          "degree": 2
        },
        "mean": 0.004211013438180089
      },
      {
        "cfg": {
          "lr": 0.006,
          "degree": 2
        },
        "mean": 0.0017769129190128297
      },
      {
        "cfg": {
          "lr": 0.001,
          "degree": 3
        },
        "mean": 0.02138834842480719
      },
      {
        "cfg": {
          "lr": 0.003,
          "degree": 3
        },
        "mean": 0.005644599266815931
      },
      {
        "cfg": {
          "lr": 0.006,
          "degree": 3
        },
        "mean": 0.0025558681518305093
      }
    ],
    "full": {
      "mean": 0.0015814606886124238,
      "std": 0.0004392499281827401,
      "per_seed": [
        0.0010742004960775375,
        0.0021969457156956196,
        0.0015373978530988097,
        0.002299107611179352,
        0.001275557209737599,
        0.0012933816760778427,
        0.0011743184877559543,
        0.0018007764592766762
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.0021942039020359516,
    "std": 0.0007557853669569652,
    "per_seed": [
      0.0022852420806884766,
      0.002649907488375902,
      0.001380756264552474,
      0.003865908132866025,
      0.001558732008561492,
      0.0023790807463228703,
      0.0016722833970561624,
      0.0017617210978642106
    ],
    "n": 8,
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "threshold": 0.15
        },
        "mean": 0.02138834842480719
      },
      {
        "cfg": {
          "lr": 0.001,
          "threshold": 0.25
        },
        "mean": 0.021395483752712607
      },
      {
        "cfg": {
          "lr": 0.001,
          "threshold": 0.4
        },
        "mean": 0.02141962433233857
      },
      {
        "cfg": {
          "lr": 0.003,
          "threshold": 0.15
        },
        "mean": 0.005644599266815931
      },
      {
        "cfg": {
          "lr": 0.003,
          "threshold": 0.25
        },
        "mean": 0.005646273668389767
      },
      {
        "cfg": {
          "lr": 0.003,
          "threshold": 0.4
        },
        "mean": 0.005661312199663371
      },
      {
        "cfg": {
          "lr": 0.006,
          "threshold": 0.15
        },
        "mean": 0.0025558681518305093
      },
      {
        "cfg": {
          "lr": 0.006,
          "threshold": 0.25
        },
        "mean": 0.0025564142561051995
      },
      {
        "cfg": {
          "lr": 0.006,
          "threshold": 0.4
        },
        "mean": 0.0025454534916207194
      }
    ]
  },
  "comparison": {
    "delta_mean": 0.0006127432134235278,
    "idea_wins": 2,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.001211041584610939,
      0.0004529617726802826,
      -0.0001566415885463357,
      0.0015668005216866732,
      0.00028317479882389307,
      0.0010856990702450275,
      0.0004979649093002081,
      -3.905536141246557e-05
    ],
    "p_value": 0.0324,
    "mde": 0.0005150508833037192,
    "mde_rel_pct": 32.5680484511838,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "mechanism_signature": {
    "confirmed": true,
    "prediction": "cubic residual proxy scales as amplitude^3",
    "observed_mean_proxy_ratio_at_1p5": 3.375,
    "mean_rho": 0.8967330753803253,
    "mean_active_fraction": 0.9953124970197678,
    "track_match": "dynamics/control pendulum; trained GRU systems evaluated on standard test MSE"
  },
  "protocol_notes": "8 paired seeds; baseline and idea share LiftedGRU; 12 epochs, batch 128; all tried learning rates are in both grids."
}