{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01,
      "weight_decay": 0.0001
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "weight_decay": 0.0
        },
        "mean": 0.1634710617363453
      },
      {
        "cfg": {
          "lr": 0.001,
          "weight_decay": 0.0001
        },
        "mean": 0.16360631212592125
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0
        },
        "mean": 0.02722445083782077
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0001
        },
        "mean": 0.027167017105966806
      },
      {
        "cfg": {
          "lr": 0.01,
          "weight_decay": 0.0
        },
        "mean": 0.013583156396634877
      },
      {
        "cfg": {
          "lr": 0.01,
          "weight_decay": 0.0001
        },
        "mean": 0.013508650125004351
      }
    ],
    "full": {
      "mean": 0.010331348807085305,
      "std": 0.006469534881760542,
      "per_seed": [
        0.006849721539765596,
        0.014833156950771809,
        0.0069954125210642815,
        0.025356309488415718,
        0.009617128409445286,
        0.0034944582730531693,
        0.009336123242974281,
        0.006168480031192303
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.013433240383164957,
    "std": 0.006230608825055202,
    "per_seed": [
      0.024367019534111023,
      0.015733294188976288,
      0.013501754030585289,
      0.01728116162121296,
      0.016352977603673935,
      0.009692054241895676,
      0.00801459327340126,
      0.0025230685714632273
    ],
    "n": 8,
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "lam": 0.0001
        },
        "mean": 0.16453255899250507
      },
      {
        "cfg": {
          "lr": 0.003,
          "lam": 0.0001
        },
        "mean": 0.029194727540016174
      },
      {
        "cfg": {
          "lr": 0.01,
          "lam": 0.0001
        },
        "mean": 0.01772080734372139
      }
    ],
    "best_cfg": {
      "lr": 0.01,
      "lam": 0.0001
    }
  },
  "comparison": {
    "delta_mean": 0.0031018915760796517,
    "idea_wins": 3,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.017517297994345427,
      0.0009001372382044792,
      0.0065063415095210075,
      -0.008075147867202759,
      0.006735849194228649,
      0.006197595968842506,
      -0.001321529969573021,
      -0.0036454114597290754
    ],
    "p_value": 0.31845,
    "mde": 0.006606975042569378,
    "mde_rel_pct": 63.950749954722966,
    "verdict": "no significant win",
    "system_worked": false
  },
  "mechanism_signature": {
    "observed_increment_mse": 0.0016462175408378243,
    "predicted_increment_mse": 0.010509551968425512,
    "endpoint_test_mse": 0.013433240383164957,
    "prediction": "integral residual uses observed increments instead of noisy finite-difference derivatives",
    "confirmed": false
  }
}