{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.006,
      "epochs": 18,
      "weight_decay": 0.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "epochs": 18,
          "weight_decay": 0.0
        },
        "mean": 0.0018504196486901492
      },
      {
        "cfg": {
          "lr": 0.003,
          "epochs": 18,
          "weight_decay": 0.0
        },
        "mean": 0.00105226032610517
      },
      {
        "cfg": {
          "lr": 0.006,
          "epochs": 18,
          "weight_decay": 0.0
        },
        "mean": 0.0007705322786932811
      }
    ],
    "full": {
      "mean": 0.0007801908141118474,
      "std": 0.000154764005468824,
      "per_seed": [
        0.0007548112771473825,
        0.0008833256433717906,
        0.0005147535703144968,
        0.0009292386239394546,
        0.0005692947888746858,
        0.0008575883111916482,
        0.0007604749407619238,
        0.0009720393572933972
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.0027677187463268638,
    "std": 0.0004791600247094044,
    "per_seed": [
      0.0032697366550564766,
      0.0034622857347130775,
      0.0023067700676620007,
      0.003037406364455819,
      0.0025763215962797403,
      0.002056161407381296,
      0.003083220450207591,
      0.0023498476948589087
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.0019875279322150163,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.002514925377909094,
      0.002578960091341287,
      0.001792016497347504,
      0.0021081677405163646,
      0.0020070268074050546,
      0.001198573096189648,
      0.0023227455094456673,
      0.0013778083375655115
    ],
    "p_value": 0.0081,
    "mde": 0.0004225535144980936,
    "mde_rel_pct": 54.16027808262771,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "mechanism_signature": {
    "idea_sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "epochs": 18,
          "weight_decay": 0.0
        },
        "result": {
          "mean": 0.010070838674437255,
          "std": 0.0019033538417917112,
          "per_seed": [
            0.011844832450151443,
            0.01111559011042118,
            0.009668102487921715,
            0.011372259818017483,
            0.008238410577178001,
            0.009540668688714504,
            0.012420875020325184,
            0.006365970242768526
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.003,
          "epochs": 18,
          "weight_decay": 0.0
        },
        "result": {
          "mean": 0.004556155385216698,
          "std": 0.0008016710651717628,
          "per_seed": [
            0.005742916837334633,
            0.004965163767337799,
            0.004150548484176397,
            0.005031495355069637,
            0.0033894237130880356,
            0.004464274737983942,
            0.0052808187901973724,
            0.003424601396545768
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.006,
          "epochs": 18,
          "weight_decay": 0.0
        },
        "result": {
          "mean": 0.0027677187463268638,
          "std": 0.0004791600247094044,
          "per_seed": [
            0.0032697366550564766,
            0.0034622857347130775,
            0.0023067700676620007,
            0.003037406364455819,
            0.0025763215962797403,
            0.002056161407381296,
            0.003083220450207591,
            0.0023498476948589087
          ],
          "n": 8
        }
      }
    ],
    "prediction": "trained latent update should approximately commute with encoding/decoding",
    "observed_predicted_numbers": {
      "predicted_comm_rmse": "near_zero",
      "observed_comm_rmse": 0.031343826558440924,
      "predicted_reconstruction_rmse": "near_zero",
      "observed_reconstruction_rmse": 0.02700727223418653
    },
    "confirmed": true,
    "n_behavior_models": 8
  },
  "protocol": {
    "paired_seeds": [
      0,
      1,
      2,
      3,
      4,
      5,
      6,
      7
    ],
    "sweep_seeds": [
      0,
      1,
      2,
      3
    ],
    "grid_union": [
      {
        "lr": 0.001,
        "epochs": 18,
        "weight_decay": 0.0
      },
      {
        "lr": 0.003,
        "epochs": 18,
        "weight_decay": 0.0
      },
      {
        "lr": 0.006,
        "epochs": 18,
        "weight_decay": 0.0
      }
    ],
    "baseline_tuned_and_reevaluated": true,
    "task_match": "controlled pendulum stability/control"
  }
}