{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.003,
      "weight_decay": 0.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "weight_decay": 0.0
        },
        "mean": 0.004407020111102611
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0
        },
        "mean": 0.0013896534073865041
      }
    ],
    "full": {
      "mean": 0.0015848711991566233,
      "std": 0.0004409408404879072,
      "per_seed": [
        0.0009277964127250016,
        0.0014898052904754877,
        0.001270615728572011,
        0.0018703961977735162,
        0.0014201324665918946,
        0.001722313230857253,
        0.0014589702477678657,
        0.002518940018489957
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.0014701014006277546,
    "std": 0.0005499829937633325,
    "per_seed": [
      0.0008705478394404054,
      0.0014663755428045988,
      0.0011225328780710697,
      0.0015238631749525666,
      0.0013210882898420095,
      0.0013340504374355078,
      0.0012874765088781714,
      0.0028348765335977077
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": -0.0001147697985288687,
    "idea_wins": 7,
    "n_pairs": 8,
    "per_seed_diffs": [
      -5.7248573284596205e-05,
      -2.34297476708889e-05,
      -0.00014808285050094128,
      -0.00034653302282094955,
      -9.904417674988508e-05,
      -0.0003882627934217453,
      -0.00017149373888969421,
      0.0003159365151077509
    ],
    "p_value": 0.17025,
    "mde": 0.00018161700236817133,
    "mde_rel_pct": 11.459417173131632,
    "verdict": "no significant win",
    "system_worked": false
  },
  "mechanism_signature": {
    "prediction": "Gaussian p+=Phi(mu/sd) for finite-horizon FTLE",
    "predicted_positive_fraction": 6.103742223005928e-28,
    "observed_positive_fraction": 0.0,
    "absolute_error": 6.103742223005928e-28,
    "confirmed": true,
    "idea_sweep": [
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0,
          "rho": 0.05,
          "sigma": 0.04
        },
        "res": {
          "mean": 0.0014701014006277546,
          "std": 0.0005499829937633325,
          "per_seed": [
            0.0008705478394404054,
            0.0014663755428045988,
            0.0011225328780710697,
            0.0015238631749525666,
            0.0013210882898420095,
            0.0013340504374355078,
            0.0012874765088781714,
            0.0028348765335977077
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0,
          "rho": 0.15,
          "sigma": 0.04
        },
        "res": {
          "mean": 0.0014701014006277546,
          "std": 0.0005499829937633325,
          "per_seed": [
            0.0008705478394404054,
            0.0014663755428045988,
            0.0011225328780710697,
            0.0015238631749525666,
            0.0013210882898420095,
            0.0013340504374355078,
            0.0012874765088781714,
            0.0028348765335977077
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0,
          "rho": 0.3,
          "sigma": 0.04
        },
        "res": {
          "mean": 0.0014701014006277546,
          "std": 0.0005499829937633325,
          "per_seed": [
            0.0008705478394404054,
            0.0014663755428045988,
            0.0011225328780710697,
            0.0015238631749525666,
            0.0013210882898420095,
            0.0013340504374355078,
            0.0012874765088781714,
            0.0028348765335977077
          ],
          "n": 8
        }
      }
    ]
  },
  "official_architecture_note": "Matched GRUCell implementation exposes recurrent state for differentiable FTLE; baseline and idea share all parameters/optimizer/data."
}