{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.003
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.009041695157065988
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.004719233373180032
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.006545157753862441
      }
    ],
    "full": {
      "mean": 0.005879616655874997,
      "std": 0.0028056293033459226,
      "per_seed": [
        0.0026779656764119864,
        0.004133712034672499,
        0.008910770528018475,
        0.0031544852536171675,
        0.006136221811175346,
        0.0032570217736065388,
        0.00849937554448843,
        0.010267380625009537
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.02712346729822457,
    "std": 0.004941447293628062,
    "per_seed": [
      0.028157928958535194,
      0.02860170602798462,
      0.0332900770008564,
      0.03025132603943348,
      0.018054386600852013,
      0.02373768761754036,
      0.02242955192923546,
      0.032465074211359024
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.02124385064234957,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.025479963282123208,
      0.02446799399331212,
      0.024379306472837925,
      0.027096840785816312,
      0.011918164789676666,
      0.02048066584393382,
      0.013930176384747028,
      0.022197693586349487
    ],
    "p_value": 0.0081,
    "mde": 0.004627304642862352,
    "mde_rel_pct": 78.70078805628737,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "mechanism_signature": {
    "mechanism_signature": {
      "trained_model": "LiePoissonRNN",
      "scales": [
        0.25,
        0.5,
        1.0
      ],
      "observed_rms_effect": [
        0.007525605789851397,
        0.014394661411643028,
        0.025677648605778813
      ],
      "observed_loglog_exponent": 0.8853167356267617,
      "confirmed": false
    },
    "idea_sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "eval": {
          "mean": 0.20517639257013798,
          "std": 0.05351073177265944,
          "per_seed": [
            0.16992242634296417,
            0.199637770652771,
            0.268087774515152,
            0.29078808426856995,
            0.18213585019111633,
            0.2410622090101242,
            0.16997386515140533,
            0.11980316042900085
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "eval": {
          "mean": 0.03456594352610409,
          "std": 0.009361537439374885,
          "per_seed": [
            0.02949955314397812,
            0.028048669919371605,
            0.052656497806310654,
            0.03622344508767128,
            0.042653679847717285,
            0.03863559663295746,
            0.021713633090257645,
            0.027096472680568695
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "eval": {
          "mean": 0.02712346729822457,
          "std": 0.004941447293628062,
          "per_seed": [
            0.028157928958535194,
            0.02860170602798462,
            0.0332900770008564,
            0.03025132603943348,
            0.018054386600852013,
            0.02373768761754036,
            0.02242955192923546,
            0.032465074211359024
          ],
          "n": 8
        }
      }
    ],
    "parameter_count_baseline": 22261,
    "parameter_count_idea": 6254
  }
}