{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.001864717691205442
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.0010901302302954718
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.0006186682076076977
      }
    ],
    "full": {
      "mean": 0.00054517226817552,
      "std": 0.00020361842727135583,
      "per_seed": [
        0.0004930912400595844,
        0.0004765945195686072,
        0.0004673648509196937,
        0.0010376222198829055,
        0.000511663849465549,
        0.00027233464061282575,
        0.0005571767687797546,
        0.0005455300561152399
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.24843271635472775,
    "std": 0.12890244786167743,
    "per_seed": [
      0.09243064373731613,
      0.48731154203414917,
      0.30256471037864685,
      0.0956249013543129,
      0.23182319104671478,
      0.24627986550331116,
      0.155330628156662,
      0.376096248626709
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.24788754408655223,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.09193755249725655,
      0.48683494751458056,
      0.30209734552772716,
      0.09458727913442999,
      0.23131152719724923,
      0.24600753086269833,
      0.15477345138788223,
      0.37555071857059374
    ],
    "p_value": 0.0081,
    "mde": 0.1153001840292749,
    "mde_rel_pct": 21149.312017491255,
    "verdict": "idea worse (significant)",
    "system_worked": false,
    "permutation_pvalue": 0.0081
  },
  "idea_config": {
    "lr": 0.01,
    "lambda": 0.01
  },
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.001,
        "lambda": 0.01
      },
      "result": {
        "mean": 0.74566999822855,
        "std": 0.043829155721253926,
        "per_seed": [
          0.728725254535675,
          0.7290393114089966,
          0.78700190782547,
          0.8437849879264832,
          0.7193636894226074,
          0.7325140237808228,
          0.7265931367874146,
          0.6983376741409302
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.001,
        "lambda": 0.1
      },
      "result": {
        "mean": 0.7457223460078239,
        "std": 0.04384320703957966,
        "per_seed": [
          0.7288874387741089,
          0.729056179523468,
          0.7870328426361084,
          0.8438847064971924,
          0.719398021697998,
          0.7325379252433777,
          0.7266111373901367,
          0.6983705163002014
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.001,
        "lambda": 1.0
      },
      "result": {
        "mean": 0.745727613568306,
        "std": 0.043844610253164565,
        "per_seed": [
          0.7289036512374878,
          0.7290578484535217,
          0.7870359420776367,
          0.8438947200775146,
          0.7194015383720398,
          0.7325403690338135,
          0.7266129851341248,
          0.6983738541603088
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.003,
        "lambda": 0.01
      },
      "result": {
        "mean": 0.6492044553160667,
        "std": 0.034374889139018386,
        "per_seed": [
          0.5766242742538452,
          0.690006673336029,
          0.6899384260177612,
          0.6342498064041138,
          0.6637277007102966,
          0.655630886554718,
          0.6534223556518555,
          0.6300355195999146
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.003,
        "lambda": 0.1
      },
      "result": {
        "mean": 0.6523455232381821,
        "std": 0.03187828804668321,
        "per_seed": [
          0.5861462950706482,
          0.6904615759849548,
          0.692155659198761,
          0.6431483030319214,
          0.6647420525550842,
          0.6567747592926025,
          0.6541696190834045,
          0.6311659216880798
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.003,
        "lambda": 1.0
      },
      "result": {
        "mean": 0.6526677086949348,
        "std": 0.03162840900701376,
        "per_seed": [
          0.5871396660804749,
          0.6905071139335632,
          0.6923784017562866,
          0.6440579891204834,
          0.6648440957069397,
          0.6568903923034668,
          0.6542445421218872,
          0.631279468536377
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.01,
        "lambda": 0.01
      },
      "result": {
        "mean": 0.24843271635472775,
        "std": 0.12890244786167743,
        "per_seed": [
          0.09243064373731613,
          0.48731154203414917,
          0.30256471037864685,
          0.0956249013543129,
          0.23182319104671478,
          0.24627986550331116,
          0.155330628156662,
          0.376096248626709
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.01,
        "lambda": 0.1
      },
      "result": {
        "mean": 0.37393518909811974,
        "std": 0.08752984377791141,
        "per_seed": [
          0.31651848554611206,
          0.5420994162559509,
          0.4366860091686249,
          0.2581859529018402,
          0.36661314964294434,
          0.349292129278183,
          0.28558772802352905,
          0.43649864196777344
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.01,
        "lambda": 1.0
      },
      "result": {
        "mean": 0.39147959649562836,
        "std": 0.08067803171474866,
        "per_seed": [
          0.35244834423065186,
          0.5486297011375427,
          0.4513516426086426,
          0.28766605257987976,
          0.38437727093696594,
          0.36130663752555847,
          0.3028562068939209,
          0.4432009160518646
        ],
        "n": 8
      }
    }
  ],
  "track_justification": "Dynamics is structurally matched: controlled actuated pendulum rollout with recurrent hidden state.",
  "mechanism_signature": {
    "prediction": "data-fitted ellipsoid barrier plus projection keeps normalized hidden energy bounded",
    "baseline_mean_hidden_energy": 2.286437749862671,
    "idea_mean_normalized_energy": 9.777409553527832,
    "idea_max_normalized_energy": 22.335248947143555,
    "idea_violation_rate": 1.0,
    "predicted_max_energy": 1.0,
    "confirmed": false
  }
}