{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.00665
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.0009722943214001134
      },
      {
        "cfg": {
          "lr": 0.002
        },
        "mean": 0.0006509074009954929
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.0004595852587954141
      },
      {
        "cfg": {
          "lr": 0.004
        },
        "mean": 0.0003482927131699398
      },
      {
        "cfg": {
          "lr": 0.005
        },
        "mean": 0.00024859366021701135
      },
      {
        "cfg": {
          "lr": 0.00067
        },
        "mean": 0.0011799605708802119
      },
      {
        "cfg": {
          "lr": 0.00134
        },
        "mean": 0.0008329209085786715
      },
      {
        "cfg": {
          "lr": 0.00268
        },
        "mean": 0.0005083113646833226
      },
      {
        "cfg": {
          "lr": 0.00402
        },
        "mean": 0.0003466853668214753
      },
      {
        "cfg": {
          "lr": 0.00536
        },
        "mean": 0.00022448933304985985
      },
      {
        "cfg": {
          "lr": 0.00665
        },
        "mean": 0.00013951877372164745
      }
    ],
    "full": {
      "mean": 0.00011474007897049887,
      "std": 4.026902293849407e-05,
      "per_seed": [
        0.0001932993473019451,
        0.0001438440231140703,
        8.013794285943732e-05,
        0.0001407937816111371,
        0.00011875866039190441,
        8.25471433927305e-05,
        6.331295298878103e-05,
        9.522678010398522e-05
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.05237803841009736,
    "std": 0.008875649004641814,
    "per_seed": [
      0.04106063395738602,
      0.058054666966199875,
      0.0539846234023571,
      0.06823215633630753,
      0.05530509725213051,
      0.03805254027247429,
      0.05319170653820038,
      0.05114288255572319
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.05226329833112686,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.04086733461008407,
      0.057910822943085805,
      0.053904485459497664,
      0.06809136255469639,
      0.055186338591738604,
      0.03796999312908156,
      0.0531283935852116,
      0.051047655775619205
    ],
    "p_value": 0.0081,
    "mde": 0.007932781978214037,
    "mde_rel_pct": 6913.697506042031,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "mechanism_signature": {
    "farkas_shield": {
      "constraint_count_full": 64,
      "constraint_count_retained": 2,
      "retained_fraction": 0.03125,
      "certificate_max_residual": 0.0,
      "certificate_max_offset_violation": 0.0,
      "trained_test_samples": 400,
      "full_reduced_max_output_difference": 0.0,
      "full_reduced_decision_disagreement": 0.0,
      "predicted_zero_disagreement_observed": true,
      "confirmed": true
    }
  },
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.00665
      },
      "result": {
        "mean": 0.05238204402849078,
        "std": 0.008873981965744437,
        "per_seed": [
          0.04106547310948372,
          0.0580538772046566,
          0.0539894662797451,
          0.06823457032442093,
          0.055308111011981964,
          0.03805993124842644,
          0.053193967789411545,
          0.05115095525979996
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.0044555
      },
      "result": {
        "mean": 0.05238870158791542,
        "std": 0.008873361924843962,
        "per_seed": [
          0.04107039421796799,
          0.0580560676753521,
          0.053996019065380096,
          0.06824300438165665,
          0.055311888456344604,
          0.03807036951184273,
          0.053200963884592056,
          0.05116090551018715
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.0088445
      },
      "result": {
        "mean": 0.05237803841009736,
        "std": 0.008875649004641814,
        "per_seed": [
          0.04106063395738602,
          0.058054666966199875,
          0.0539846234023571,
          0.06823215633630753,
          0.05530509725213051,
          0.03805254027247429,
          0.05319170653820038,
          0.05114288255572319
        ],
        "n": 8
      }
    }
  ],
  "protocol_notes": {
    "task_match": "controlled pendulum dynamics; safety/control idea",
    "epochs": 15,
    "n_train": 1200,
    "n_test": 400,
    "idea_uses_same_base_architecture": true
  }
}