{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.2699683979153633
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.01333702343981713
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.009317408432252705
      }
    ],
    "full": {
      "mean": 0.009518424805719405,
      "std": 0.004154286071903827,
      "per_seed": [
        0.013998116366565228,
        0.005631609819829464,
        0.013994917273521423,
        0.0036449902690947056,
        0.007559004705399275,
        0.006567755248397589,
        0.00925444532185793,
        0.01549655944108963
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.2998440582305193,
    "std": 0.07520865063647589,
    "per_seed": [
      0.39651888608932495,
      0.2933981418609619,
      0.41620975732803345,
      0.24436108767986298,
      0.27874669432640076,
      0.17241966724395752,
      0.3343149721622467,
      0.2627832591533661
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.2903256334247999,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.3825207697227597,
      0.28776653204113245,
      0.402214840054512,
      0.24071609741076827,
      0.2711876896210015,
      0.16585191199555993,
      0.3250605268403888,
      0.24728669971227646
    ],
    "p_value": 0.0081,
    "mde": 0.0650801969908384,
    "mde_rel_pct": 683.7286454344126,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "mechanism_signature": {
    "prediction": "trained averaged/fast state error decreases with eps; mu2 is negative",
    "rows": [
      {
        "eps": 0.5,
        "rms_state_error": 0.0011420742375776172
      },
      {
        "eps": 0.25,
        "rms_state_error": 0.0007367426878772676
      },
      {
        "eps": 0.125,
        "rms_state_error": 0.0004585969727486372
      },
      {
        "eps": 0.0625,
        "rms_state_error": 0.0007371138781309128
      }
    ],
    "observed_loglog_slope": 0.2579025020059939,
    "predicted_error_slope": 1.0,
    "predicted_mu2_upper_bound": -1.35,
    "observed_max_mu2": -1.5680792331695557,
    "confirmed": false
  },
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.001
      },
      "result": {
        "mean": 0.7121945843100548,
        "std": 0.0857891837384087,
        "per_seed": [
          0.8319963812828064,
          0.690710723400116,
          0.7957785129547119,
          0.7097331285476685,
          0.6804750561714172,
          0.5556079149246216,
          0.7924796342849731,
          0.6407753229141235
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.003
      },
      "result": {
        "mean": 0.6083195172250271,
        "std": 0.0816738853448453,
        "per_seed": [
          0.7198447585105896,
          0.5920082926750183,
          0.7039626836776733,
          0.5942644476890564,
          0.5862480998039246,
          0.4565565288066864,
          0.6692549586296082,
          0.5444163680076599
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.01
      },
      "result": {
        "mean": 0.2998440582305193,
        "std": 0.07520865063647589,
        "per_seed": [
          0.39651888608932495,
          0.2933981418609619,
          0.41620975732803345,
          0.24436108767986298,
          0.27874669432640076,
          0.17241966724395752,
          0.3343149721622467,
          0.2627832591533661
        ],
        "n": 8
      }
    }
  ],
  "runtime_seconds": 79.51734733581543
}