{
  "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": 0.004545392468571663
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0
        },
        "mean": 0.00358188062091358
      },
      {
        "cfg": {
          "lr": 0.01,
          "weight_decay": 0.0
        },
        "mean": 0.0015466218756046146
      }
    ],
    "full": {
      "mean": 0.0014403377863345668,
      "std": 0.0003958335789015625,
      "per_seed": [
        0.001036701025441289,
        0.0019527574768289924,
        0.0010734819807112217,
        0.0021235470194369555,
        0.001080187619663775,
        0.0011896038195118308,
        0.001624842407181859,
        0.001441580941900611
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.004763860866660252,
    "std": 0.002933690813628907,
    "per_seed": [
      0.011456220410764217,
      0.002699522068724036,
      0.0031392518430948257,
      0.0025344439782202244,
      0.007165998220443726,
      0.0028000844176858664,
      0.003200623905286193,
      0.005114742089062929
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.0033235230803256854,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.010419519385322928,
      0.0007467645918950438,
      0.002065769862383604,
      0.00041089695878326893,
      0.006085810600779951,
      0.0016104805981740355,
      0.0015757814981043339,
      0.0036731611471623182
    ],
    "p_value": 0.0081,
    "mde": 0.002840810147756273,
    "mde_rel_pct": 197.2322169638893,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "mechanism_signature": {
    "idea_sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "weight_decay": 0.0
        },
        "result": {
          "mean": 0.004763860866660252,
          "std": 0.002933690813628907,
          "per_seed": [
            0.011456220410764217,
            0.002699522068724036,
            0.0031392518430948257,
            0.0025344439782202244,
            0.007165998220443726,
            0.0028000844176858664,
            0.003200623905286193,
            0.005114742089062929
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0
        },
        "result": {
          "mean": 0.0048263132775900885,
          "std": 0.0025504190600527878,
          "per_seed": [
            0.0045633199624717236,
            0.004938819445669651,
            0.0009674743050709367,
            0.010774293914437294,
            0.004922726657241583,
            0.004287341609597206,
            0.0037127339746803045,
            0.00444379635155201
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.01,
          "weight_decay": 0.0
        },
        "result": {
          "mean": 0.04266263055615127,
          "std": 0.02664842856440231,
          "per_seed": [
            0.07184934616088867,
            0.04949978366494179,
            0.023517103865742683,
            0.029628023505210876,
            0.008588656783103943,
            0.09184401482343674,
            0.048785608261823654,
            0.017588507384061813
          ],
          "n": 8
        }
      }
    ],
    "selected_cfg": {
      "lr": 0.001,
      "weight_decay": 0.0
    },
    "signature": {
      "prediction": "latent box transitions should contain observed trained-model successors",
      "observed_transitions": 159,
      "same_cell_fraction": 0.0,
      "mean_observed_box_span": 0.17268966138362885,
      "cells": 4,
      "metric_for_signature_model": 0.011456220410764217,
      "confirmed": true
    }
  }
}