{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.006
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.0015
        },
        "mean": 0.0011108634498668835
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.0008105679735308513
      },
      {
        "cfg": {
          "lr": 0.006
        },
        "mean": 0.000628942609182559
      }
    ],
    "full": {
      "mean": 0.0005640739254886284,
      "std": 0.0001367984084942288,
      "per_seed": [
        0.00048537924885749817,
        0.0007160367094911635,
        0.000499650021083653,
        0.0008147044572979212,
        0.0004540993832051754,
        0.0005191828240640461,
        0.0003816201351583004,
        0.0006419186247512698
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.0005420081324700732,
    "std": 0.00013960018191357758,
    "per_seed": [
      0.000509127858094871,
      0.0005584119935519993,
      0.000469617429189384,
      0.0008545061573386192,
      0.00044199760304763913,
      0.0003986096999142319,
      0.00044967373833060265,
      0.0006541205802932382
    ],
    "n": 8,
    "best_cfg": {
      "lr": 0.006,
      "alpha": 2.0,
      "beta": 0.9,
      "sigma": 0.05
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.0015,
          "alpha": 1.0,
          "beta": 0.9,
          "sigma": 0.05
        },
        "mean": 0.0013087970073684119
      },
      {
        "cfg": {
          "lr": 0.003,
          "alpha": 1.5,
          "beta": 0.9,
          "sigma": 0.05
        },
        "mean": 0.0008830113292788155
      },
      {
        "cfg": {
          "lr": 0.006,
          "alpha": 2.0,
          "beta": 0.9,
          "sigma": 0.05
        },
        "mean": 0.0005420081324700732
      }
    ]
  },
  "comparison": {
    "delta_mean": -2.206579301855527e-05,
    "idea_wins": 4,
    "n_pairs": 8,
    "per_seed_diffs": [
      2.3748609237372875e-05,
      -0.00015762471593916416,
      -3.003259189426899e-05,
      3.980170004069805e-05,
      -1.2101780157536268e-05,
      -0.00012057312414981425,
      6.805360317230225e-05,
      1.2201955541968346e-05
    ],
    "p_value": 0.44995,
    "mde": 6.588809659856915e-05,
    "mde_rel_pct": 11.680755592716622,
    "verdict": "no significant win",
    "system_worked": false
  },
  "protocol_notes": {
    "epochs": 8,
    "n_train": 1200,
    "n_test": 400,
    "batch": 128,
    "track_choice": "dynamics matches stability/control and multi-step pendulum rollouts",
    "idea": "residual-adaptive candidate trajectory sampling",
    "lr_union": [
      0.0015,
      0.003,
      0.006
    ]
  },
  "mechanism_signature": {
    "prediction": "larger trained-model observed residual induces larger temperature and flatter candidate weights",
    "observed_residual_mean": 0.01952467652328778,
    "observed_residual_p90": 0.03243881464004517,
    "observed_ESS_low_residual": 25.594260510103627,
    "observed_ESS_high_residual": 28.17312607209512,
    "ESS_ratio_high_over_low": 1.1007595261825773,
    "confirmed": true
  }
}