{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01,
      "weight_decay": 0.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "weight_decay": 0.0
        },
        "mean": 0.001154391880845651
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0
        },
        "mean": 0.0006730457389494404
      },
      {
        "cfg": {
          "lr": 0.01,
          "weight_decay": 0.0
        },
        "mean": 0.00024367208061448764
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0001
        },
        "mean": 0.00066975416848436
      }
    ],
    "full": {
      "mean": 0.00020419583597686142,
      "std": 0.00010942496465474286,
      "per_seed": [
        0.00037123056245036423,
        0.00033986172638833523,
        7.115668995538726e-05,
        0.00019243934366386384,
        0.00017716105503495783,
        0.00011334212467772886,
        7.963812095113099e-05,
        0.0002887370646931231
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.3719000369310379,
    "std": 0.1780352000875736,
    "per_seed": [
      0.25049954652786255,
      0.32161659002304077,
      0.2485007792711258,
      0.6178863644599915,
      0.1744787096977234,
      0.22901414334774017,
      0.45399707555770874,
      0.6792070865631104
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.37169584109506104,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.2501283159654122,
      0.32127672829665244,
      0.2484296225811704,
      0.6176939251163276,
      0.17430154864268843,
      0.22890080122306244,
      0.4539174374367576,
      0.6789183494984172
    ],
    "p_value": 0.0081,
    "mde": 0.1591303337139686,
    "mde_rel_pct": 77930.25403906891,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "mechanism_signature": {
    "mechanism_signature": {
      "prediction": "compatible projection reduces unsupported hidden transition energy/rank without changing task architecture",
      "observed": [
        {
          "seed": 0,
          "metric": 0.25049954652786255,
          "projection_mse": 0.013389366678893566,
          "rank": 2,
          "raw_spectral_radius": 1.101452112197876
        },
        {
          "seed": 1,
          "metric": 0.32161659002304077,
          "projection_mse": 0.013996558263897896,
          "rank": 1,
          "raw_spectral_radius": 1.0808608531951904
        },
        {
          "seed": 2,
          "metric": 0.2485007792711258,
          "projection_mse": 0.011121638119220734,
          "rank": 2,
          "raw_spectral_radius": 1.0855185985565186
        },
        {
          "seed": 3,
          "metric": 0.6178863644599915,
          "projection_mse": 0.018192939460277557,
          "rank": 2,
          "raw_spectral_radius": 1.1126620769500732
        }
      ],
      "mean_projection_mse": 0.014175125630572438,
      "mean_rank": 1.75,
      "confirmed": true
    },
    "custom_track": null,
    "idea_cfg": {
      "lr": 0.01,
      "weight_decay": 0.0,
      "tau": 0.15,
      "levels": 1
    },
    "idea_grid": [
      {
        "lr": 0.01,
        "weight_decay": 0.0,
        "tau": 0.1,
        "levels": 1
      },
      {
        "lr": 0.01,
        "weight_decay": 0.0,
        "tau": 0.15,
        "levels": 1
      },
      {
        "lr": 0.01,
        "weight_decay": 0.0,
        "tau": 0.25,
        "levels": 1
      }
    ]
  }
}