{
  "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": 0.001
        },
        "mean": 0.0005152521189302206
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 4.9621185098658316e-05
      },
      {
        "cfg": {
          "lr": 0.006
        },
        "mean": 2.345215943932999e-05
      }
    ],
    "full": {
      "mean": 2.282289460708853e-05,
      "std": 2.305330446600017e-06,
      "per_seed": [
        2.4922346710809506e-05,
        2.6788266040966846e-05,
        1.934224746946711e-05,
        2.2755777536076494e-05,
        2.3468126528314315e-05,
        2.385292282269802e-05,
        2.0721481632790528e-05,
        2.073198811558541e-05
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 2.4683711671968922e-05,
    "std": 2.8119869264952184e-06,
    "per_seed": [
      2.5581410227459855e-05,
      2.4755079721217044e-05,
      1.922530645970255e-05,
      2.4516248231520876e-05,
      2.3418831915478222e-05,
      2.4992359612951986e-05,
      2.4735531042097136e-05,
      3.0244926165323704e-05
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 1.8608170648803934e-06,
    "idea_wins": 3,
    "n_pairs": 8,
    "per_seed_diffs": [
      6.590635166503489e-07,
      -2.0331863197498024e-06,
      -1.1694100976455957e-07,
      1.7604706954443827e-06,
      -4.929461283609271e-08,
      1.139436790253967e-06,
      4.014049409306608e-06,
      9.512938049738295e-06
    ],
    "p_value": 0.1888,
    "mde": 2.9628199598181274e-06,
    "mde_rel_pct": 12.981788729366123,
    "verdict": "no significant win",
    "system_worked": false
  },
  "mechanism_signature": {
    "math_sanity": {
      "stationary_target_alpha": 0.2,
      "stationary_observed_exceedance": 0.20028,
      "stationary_abs_error": 0.00028000000000000247,
      "ramp_observed_steps": 19,
      "ramp_predicted_steps": 19,
      "recovery_observed_steps": 88,
      "recovery_predicted_steps": 88,
      "quantile_check": 0.8500000000000001
    },
    "trained_model_signature": {
      "test_abs_error_mean": 0.003862829878926277,
      "test_abs_error_q90": 0.007725757360458375,
      "retested_exceedance": 0.197,
      "target_alpha": 0.2,
      "final_radius": 0.0,
      "confirmed": true,
      "radius_training_start_end": [
        0.19,
        0.0
      ]
    },
    "intervention": "adaptive conformal exceedance-weighted MSE",
    "structural_match": "controlled pendulum multi-step dynamics"
  },
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.001,
        "alpha": 0.2,
        "gamma": 0.05,
        "strength": 0.5
      },
      "result": {
        "mean": 0.0004968880639353301,
        "std": 4.8528307616895607e-05,
        "per_seed": [
          0.0004283393791411072,
          0.0005958400433883071,
          0.0004622905398719013,
          0.0005341034266166389,
          0.0004780906019732356,
          0.0004924809327349067,
          0.000466031429823488,
          0.0005179281579330564
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.003,
        "alpha": 0.2,
        "gamma": 0.05,
        "strength": 0.5
      },
      "result": {
        "mean": 5.3715010380983585e-05,
        "std": 1.4501156678207892e-05,
        "per_seed": [
          8.585918840253726e-05,
          6.12151634413749e-05,
          4.8918511311057955e-05,
          3.291441680630669e-05,
          5.7153552916133776e-05,
          4.41737865912728e-05,
          4.8585090553388e-05,
          5.090037302579731e-05
        ],
        "n": 8
      }
    },
    {
      "cfg": {
        "lr": 0.006,
        "alpha": 0.2,
        "gamma": 0.05,
        "strength": 0.5
      },
      "result": {
        "mean": 2.4683711671968922e-05,
        "std": 2.8119869264952184e-06,
        "per_seed": [
          2.5581410227459855e-05,
          2.4755079721217044e-05,
          1.922530645970255e-05,
          2.4516248231520876e-05,
          2.3418831915478222e-05,
          2.4992359612951986e-05,
          2.4735531042097136e-05,
          3.0244926165323704e-05
        ],
        "n": 8
      }
    }
  ]
}