{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.006
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.006104337517172098
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.0017399389762431383
      },
      {
        "cfg": {
          "lr": 0.006
        },
        "mean": 0.0012507251958595589
      }
    ],
    "full": {
      "mean": 0.001170862014987506,
      "std": 0.0004510333101838233,
      "per_seed": [
        0.000902677362319082,
        0.0019194456981495023,
        0.0005950153572484851,
        0.0015857623657211661,
        0.0008504731813445687,
        0.000836688035633415,
        0.0016707839677110314,
        0.001006050151772797
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.0011421147792134434,
    "std": 0.00044998942636985775,
    "per_seed": [
      0.0008761535282246768,
      0.0019359904108569026,
      0.0006074620177969337,
      0.0015396701637655497,
      0.0007932198350317776,
      0.0007940693758428097,
      0.0015995833091437817,
      0.0009907695930451155
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": -2.8747235774062574e-05,
    "idea_wins": 6,
    "n_pairs": 8,
    "per_seed_diffs": [
      -2.6523834094405174e-05,
      1.6544712707400322e-05,
      1.2446660548448563e-05,
      -4.6092201955616474e-05,
      -5.725334631279111e-05,
      -4.261865979060531e-05,
      -7.120065856724977e-05,
      -1.5280558727681637e-05
    ],
    "p_value": 0.05535,
    "mde": 2.6522118750884212e-05,
    "mde_rel_pct": 2.265178852109847,
    "verdict": "no significant win",
    "system_worked": false
  },
  "mechanism_signature": {
    "mechanism_signature": {
      "mean_residual": 0.6372732743620872,
      "retained_modes": 48.0,
      "retained_fraction": 0.75,
      "fit_rmse": 0.015746959019452333,
      "latent_step_rms": 0.16071157343685627,
      "predicted_vs_observed": {
        "predicted_retained_fraction": 0.75,
        "observed_retained_fraction": 0.75,
        "prediction": "screening retains approximately the lower 75% residual modes"
      },
      "confirmed": true
    },
    "idea_lr_runs": {
      "0.001": {
        "mean": 0.005698318331269547,
        "std": 0.0013604234038858917,
        "per_seed": [
          0.00519864447414875,
          0.008421543054282665,
          0.005426613613963127,
          0.005435609724372625,
          0.00428318977355957,
          0.0038333924021571875,
          0.006848391145467758,
          0.006139162462204695
        ],
        "n": 8
      },
      "0.003": {
        "mean": 0.0016669390752213076,
        "std": 0.0002611493893328423,
        "per_seed": [
          0.002052778610959649,
          0.0019236018415540457,
          0.0013689034385606647,
          0.0015125471400097013,
          0.0015859621344134212,
          0.0015673021553084254,
          0.0019797401037067175,
          0.0013446771772578359
        ],
        "n": 8
      },
      "0.006": {
        "mean": 0.0011421147792134434,
        "std": 0.00044998942636985775,
        "per_seed": [
          0.0008761535282246768,
          0.0019359904108569026,
          0.0006074620177969337,
          0.0015396701637655497,
          0.0007932198350317776,
          0.0007940693758428097,
          0.0015995833091437817,
          0.0009907695930451155
        ],
        "n": 8
      }
    },
    "architecture_match": "same bench rnn_small GRU and head; only residual loss differs",
    "alpha": 0.03,
    "quantile": 0.75
  },
  "protocol_notes": {
    "baseline_grid": [
      0.001,
      0.003,
      0.006
    ],
    "idea_grid": [
      0.001,
      0.003,
      0.006
    ],
    "epochs": 15,
    "batch": 128,
    "n_train": 400,
    "n_test": 400,
    "selection": "baseline selected by four-seed sweep; idea best selected among same grid",
    "structural_match": "controlled pendulum dynamics requires temporal latent stability"
  }
}