{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.006,
      "weight_decay": 0.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.0015,
          "weight_decay": 0.0
        },
        "mean": 0.0017880297382362187
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0
        },
        "mean": 0.0012683377135545015
      },
      {
        "cfg": {
          "lr": 0.006,
          "weight_decay": 0.0
        },
        "mean": 0.0008017216168809682
      }
    ],
    "full": {
      "mean": 0.0007909468331490643,
      "std": 0.00016569300329491533,
      "per_seed": [
        0.0008831982268020511,
        0.0008114342344924808,
        0.0005164801841601729,
        0.000995773822069168,
        0.0005416677449829876,
        0.0009393909713253379,
        0.0007582727121189237,
        0.0008813567692413926
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.0008200057927751914,
    "std": 0.0001585617147958542,
    "per_seed": [
      0.0008831982268020511,
      0.0008114342344924808,
      0.0005565483006648719,
      0.0009936458664014935,
      0.0005761774955317378,
      0.0009393909713253379,
      0.0008254630374722183,
      0.0009741882095113397
    ],
    "n": 8,
    "best_cfg": {
      "lr": 0.006,
      "weight_decay": 0.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.0015,
          "weight_decay": 0.0
        },
        "mean": 0.00219198341073934,
        "per_seed": [
          0.0010475046001374722,
          0.0022253927309066057,
          0.0021842154674232006,
          0.00284200394526124,
          0.001567519037052989,
          0.001590907690115273,
          0.0031952967401593924,
          0.0028830270748585463
        ]
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0
        },
        "mean": 0.001621349343622569,
        "per_seed": [
          0.0009604159276932478,
          0.0015390310436487198,
          0.0013559971703216434,
          0.0021366437431424856,
          0.0009070083615370095,
          0.00200911913998425,
          0.0018186086090281606,
          0.0022439707536250353
        ]
      },
      {
        "cfg": {
          "lr": 0.006,
          "weight_decay": 0.0
        },
        "mean": 0.0008200057927751914,
        "per_seed": [
          0.0008831982268020511,
          0.0008114342344924808,
          0.0005565483006648719,
          0.0009936458664014935,
          0.0005761774955317378,
          0.0009393909713253379,
          0.0008254630374722183,
          0.0009741882095113397
        ]
      }
    ]
  },
  "comparison": {
    "delta_mean": 2.9058959626127034e-05,
    "idea_wins": 1,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.0,
      0.0,
      4.006811650469899e-05,
      -2.1279556676745415e-06,
      3.450975054875016e-05,
      0.0,
      6.719032535329461e-05,
      9.283144026994705e-05
    ],
    "p_value": 0.12595,
    "mde": 3.027644738730695e-05,
    "mde_rel_pct": 3.8278738997872637,
    "verdict": "no significant win",
    "system_worked": false
  },
  "mechanism_signature": {
    "math_check": {
      "rows": [
        {
          "a": 0.2,
          "predicted_variance": 1.0416666666666667,
          "observed_variance": 1.0342422123173947,
          "predicted_rho": 0.2,
          "observed_rho": 0.20071559852735493,
          "variance_relative_error": 0.007127476175301197,
          "rho_abs_error": 0.0007155985273549215
        },
        {
          "a": 0.5,
          "predicted_variance": 1.3333333333333333,
          "observed_variance": 1.3261153307140452,
          "predicted_rho": 0.5,
          "observed_rho": 0.5016303562384845,
          "variance_relative_error": 0.00541350196446605,
          "rho_abs_error": 0.0016303562384845094
        },
        {
          "a": 0.7,
          "predicted_variance": 1.9607843137254901,
          "observed_variance": 1.9111379220646099,
          "predicted_rho": 0.7,
          "observed_rho": 0.6959265998902273,
          "variance_relative_error": 0.02531965974704893,
          "rho_abs_error": 0.00407340010977264
        },
        {
          "a": 0.85,
          "predicted_variance": 3.6036036036036028,
          "observed_variance": 3.595122775671891,
          "predicted_rho": 0.85,
          "observed_rho": 0.8512361072857468,
          "variance_relative_error": 0.0023534297510499906,
          "rho_abs_error": 0.001236107285746857
        },
        {
          "a": 0.93,
          "predicted_variance": 7.401924500370103,
          "observed_variance": 7.427471227984546,
          "predicted_rho": 0.93,
          "observed_rho": 0.9301212269556333,
          "variance_relative_error": 0.0034513629007112147,
          "rho_abs_error": 0.0001212269556332135
        }
      ],
      "max_variance_relative_error": 0.02531965974704893,
      "max_rho_abs_error": 0.00407340010977264
    },
    "observable": "trained rnn_small minibatch gradient norm",
    "prediction": "joint high AR(1) and rising variance precede LR intervention",
    "observed": [
      {
        "seed": 0,
        "max_a_hat": 0.7482613680202572,
        "variance_slope": -0.0044005689645504,
        "interventions": 0
      },
      {
        "seed": 1,
        "max_a_hat": 0.7772662940130003,
        "variance_slope": -0.0034576642656289226,
        "interventions": 0
      },
      {
        "seed": 2,
        "max_a_hat": 0.7882554383146971,
        "variance_slope": -0.002863924976655482,
        "interventions": 1
      },
      {
        "seed": 3,
        "max_a_hat": 0.7757743698679345,
        "variance_slope": -0.003068314824752291,
        "interventions": 1
      },
      {
        "seed": 4,
        "max_a_hat": 0.8173354197062572,
        "variance_slope": -0.0021798070528012223,
        "interventions": 3
      },
      {
        "seed": 5,
        "max_a_hat": 0.7417832255136816,
        "variance_slope": -0.003818343265481793,
        "interventions": 0
      },
      {
        "seed": 6,
        "max_a_hat": 0.8150579419646545,
        "variance_slope": -0.0023863047120519687,
        "interventions": 1
      },
      {
        "seed": 7,
        "max_a_hat": 1.0173393024141661,
        "variance_slope": -0.0019478986432107611,
        "interventions": 12
      }
    ],
    "mean_max_a_hat": 0.810134169976831,
    "mean_variance_slope": -0.003015353338141605,
    "confirmed": false
  },
  "protocol_notes": {
    "paired_seeds": [
      0,
      1,
      2,
      3,
      4,
      5,
      6,
      7
    ],
    "dataset_sizes": [
      600,
      240
    ],
    "idea_lr_union": [
      0.0015,
      0.003,
      0.006
    ],
    "baseline_lr_union": [
      0.0015,
      0.003,
      0.006
    ],
    "track_reason": "dynamics structurally matches stability/control monitor"
  }
}