{
  "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": 0.0061489523795899
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.0013683813158422709
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.0007866571540944278
      }
    ],
    "full": {
      "mean": 0.0006869791614008136,
      "std": 0.0001962519169577577,
      "per_seed": [
        0.0008465873543173075,
        0.0005530674243345857,
        0.0006530178943648934,
        0.0010939559433609247,
        0.0006159734330140054,
        0.0003910730010829866,
        0.0006219872157089412,
        0.0007201710250228643
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.000751409461372532,
    "std": 0.0002372904792985927,
    "per_seed": [
      0.001134159043431282,
      0.0006296231294982135,
      0.0008904425776563585,
      0.0010553350439295173,
      0.0005659241578541696,
      0.0003978781751357019,
      0.0006218505441211164,
      0.0007160630193538964
    ],
    "n": 8,
    "best_cfg": {
      "lr": 0.01
    }
  },
  "comparison": {
    "delta_mean": 6.443029997171834e-05,
    "idea_wins": 4,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.00028757168911397457,
      7.655570516362786e-05,
      0.00023742468329146504,
      -3.862089943140745e-05,
      -5.0049275159835815e-05,
      6.8051740527153015e-06,
      -1.3667158782482147e-07,
      -4.108005668967962e-06
    ],
    "p_value": 0.2782,
    "mde": 0.00010753523221065781,
    "mde_rel_pct": 15.6533470376865,
    "verdict": "no significant win",
    "system_worked": false
  },
  "mechanism_signature": {
    "low_q25_sigma": 0.07731843180954456,
    "high_q75_sigma": 0.1157134473323822,
    "low_excitation_train_mse": 0.0012986520305275917,
    "high_excitation_train_mse": 0.0005935525405220687,
    "observed_error_gap_low_minus_high": 0.000705099490005523,
    "prediction": "low-excitation windows have higher residual error",
    "confirmed": true
  },
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.001
      },
      "mean": 0.00511605893552769,
      "std": 0.0024142004953189313,
      "per_seed": [
        0.0015621563652530313,
        0.005496674682945013,
        0.007589519023895264,
        0.006733444053679705,
        0.0014989599585533142,
        0.0036795316264033318,
        0.0063422685489058495,
        0.00802591722458601
      ],
      "n": 8
    },
    {
      "cfg": {
        "lr": 0.003
      },
      "mean": 0.001308875231188722,
      "std": 0.00023694831535078266,
      "per_seed": [
        0.0010585704585537314,
        0.0011665996862575412,
        0.001426042872481048,
        0.0014445277629420161,
        0.0009118119487538934,
        0.0012852930231019855,
        0.0014918958768248558,
        0.0016862602205947042
      ],
      "n": 8
    },
    {
      "cfg": {
        "lr": 0.01
      },
      "mean": 0.000751409461372532,
      "std": 0.0002372904792985927,
      "per_seed": [
        0.001134159043431282,
        0.0006296231294982135,
        0.0008904425776563585,
        0.0010553350439295173,
        0.0005659241578541696,
        0.0003978781751357019,
        0.0006218505441211164,
        0.0007160630193538964
      ],
      "n": 8
    }
  ],
  "method": {
    "description": "Per-window latent-frame consistency loss weighted by normalized smallest Jacobian singular value.",
    "tau": 0.2,
    "floor": 0.15,
    "epochs": 15,
    "batch": 128
  },
  "protocol_notes": "Matched dynamics track: controlled pendulum rollout and recurrent GRU. Baseline and idea share rnn_small, Adam, epochs, batch, data, and the union learning-rate grid; idea was evaluated at baseline-best and two nearby settings."
}