{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.006,
      "weight_decay": 0.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.0015,
          "weight_decay": 0.0
        },
        "mean": 0.08054963732138276
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0
        },
        "mean": 0.023347282083705068
      },
      {
        "cfg": {
          "lr": 0.006,
          "weight_decay": 0.0
        },
        "mean": 0.008135722251608968
      }
    ],
    "full": {
      "mean": 0.006979767291340977,
      "std": 0.0037932412616514086,
      "per_seed": [
        0.0038006319664418697,
        0.0048443530686199665,
        0.007569973357021809,
        0.016327930614352226,
        0.004687489941716194,
        0.004746445454657078,
        0.008039474487304688,
        0.005821839440613985
      ],
      "n": 8
    }
  },
  "idea": {
    "per_seed": [
      0.003616119036450982,
      0.0048443530686199665,
      0.007523232605308294,
      0.015394117683172226,
      0.0045921714045107365,
      0.004602333996444941,
      0.00775922043249011,
      0.005821839440613985
    ],
    "mean": 0.006769173458451405
  },
  "comparison": {
    "delta_mean": -0.00021059383288957179,
    "idea_wins": 6,
    "n_pairs": 8,
    "per_seed_diffs": [
      -0.00018451292999088764,
      0.0,
      -4.674075171351433e-05,
      -0.0009338129311800003,
      -9.531853720545769e-05,
      -0.00014411145821213722,
      -0.0002802540548145771,
      0.0
    ],
    "p_value": 0.0311,
    "mde": 0.00025710573309478267,
    "mde_rel_pct": 3.683586033215538,
    "verdict": "idea better (significant)",
    "system_worked": true
  },
  "mechanism_signature": {
    "window": 8,
    "predicted_margin_mean": 0.8796176314353943,
    "observed_input_sensitivity_mean": 0.0076210699044167995,
    "predicted_vs_observed_correlation": 0.19263989726460348,
    "confirmed": false
  },
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.0015,
        "weight_decay": 0.0
      },
      "mean": 0.06897093774750829
    },
    {
      "cfg": {
        "lr": 0.003,
        "weight_decay": 0.0
      },
      "mean": 0.02178938896395266
    },
    {
      "cfg": {
        "lr": 0.006,
        "weight_decay": 0.0
      },
      "mean": 0.007844455598387867
    }
  ],
  "protocol_notes": "8 paired seeds; baseline sweep seeds 0-3 plus full best-config evaluation; idea uses same three learning rates and epochs."
}