{
  "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": 0.0018504196486901492
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.00105226032610517
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.0005928000318817794
      }
    ],
    "full": {
      "per_seed": [
        0.000473822612548247,
        0.0005578889395110309,
        0.00041418071486987174,
        0.0009253078605979681,
        0.0004536386113613844,
        0.0002683700295165181,
        0.0006777027156203985,
        0.000613455253187567
      ],
      "mean": 0.0005480458421516232,
      "std": 0.00019787191330725143
    }
  },
  "idea": {
    "per_seed": [
      0.06354859471321106,
      0.07997369021177292,
      0.08213100582361221,
      0.06680075824260712,
      0.07302123308181763,
      0.06840986758470535,
      0.06384869664907455,
      0.07170670479536057
    ],
    "mean": 0.07118006888777018,
    "std": 0.006966848239001737,
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "kappa": 0.25
        },
        "mean": 0.07311351224780083
      },
      {
        "cfg": {
          "lr": 0.001,
          "kappa": 0.5
        },
        "mean": 0.11853109858930111
      },
      {
        "cfg": {
          "lr": 0.001,
          "kappa": 1.0
        },
        "mean": 0.24801216274499893
      },
      {
        "cfg": {
          "lr": 0.003,
          "kappa": 0.25
        },
        "mean": 0.08233587071299553
      },
      {
        "cfg": {
          "lr": 0.003,
          "kappa": 0.5
        },
        "mean": 0.10741026885807514
      },
      {
        "cfg": {
          "lr": 0.003,
          "kappa": 1.0
        },
        "mean": 0.17653513327240944
      },
      {
        "cfg": {
          "lr": 0.01,
          "kappa": 0.25
        },
        "mean": 0.0779394619166851
      },
      {
        "cfg": {
          "lr": 0.01,
          "kappa": 0.5
        },
        "mean": 0.10071836225688457
      },
      {
        "cfg": {
          "lr": 0.01,
          "kappa": 1.0
        },
        "mean": 0.23662623763084412
      }
    ],
    "best_cfg": {
      "lr": 0.001,
      "kappa": 0.25
    }
  },
  "comparison": {
    "delta_mean": 0.07063202304561855,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.06307477210066281,
      0.07941580127226189,
      0.08171682510874234,
      0.06587545038200915,
      0.07256759447045624,
      0.06814149755518883,
      0.06317099393345416,
      0.071093249542173
    ],
    "p_value": 0.0081,
    "mde": 0.005875162592988052,
    "mde_rel_pct": 1072.0202839095018,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "mechanism_signature": {
    "per_seed": [
      {
        "seed": 0,
        "observed_residual": 0.2520884573459625,
        "normalized_residual": 0.7100334684505998,
        "tightening_margin": 0.17750836711264995,
        "predicted_safe_fraction": 0.98,
        "observed_safe_fraction": 0.9
      },
      {
        "seed": 1,
        "observed_residual": 0.2827962040901184,
        "normalized_residual": 0.7726593654177988,
        "tightening_margin": 0.1931648413544497,
        "predicted_safe_fraction": 0.9675,
        "observed_safe_fraction": 0.865
      },
      {
        "seed": 2,
        "observed_residual": 0.2865850627422333,
        "normalized_residual": 0.7903462817844339,
        "tightening_margin": 0.19758657044610847,
        "predicted_safe_fraction": 0.98,
        "observed_safe_fraction": 0.9025
      },
      {
        "seed": 3,
        "observed_residual": 0.2584584355354309,
        "normalized_residual": 0.7446077562600646,
        "tightening_margin": 0.18615193906501615,
        "predicted_safe_fraction": 0.98,
        "observed_safe_fraction": 0.9
      },
      {
        "seed": 4,
        "observed_residual": 0.2702244222164154,
        "normalized_residual": 0.7408200912064927,
        "tightening_margin": 0.18520502280162318,
        "predicted_safe_fraction": 0.98,
        "observed_safe_fraction": 0.8825
      },
      {
        "seed": 5,
        "observed_residual": 0.2615528106689453,
        "normalized_residual": 0.7201894427650053,
        "tightening_margin": 0.18004736069125132,
        "predicted_safe_fraction": 0.995,
        "observed_safe_fraction": 0.895
      },
      {
        "seed": 6,
        "observed_residual": 0.25268301367759705,
        "normalized_residual": 0.7071742747233145,
        "tightening_margin": 0.17679356868082863,
        "predicted_safe_fraction": 0.9825,
        "observed_safe_fraction": 0.9025
      },
      {
        "seed": 7,
        "observed_residual": 0.2677810788154602,
        "normalized_residual": 0.7319731444569756,
        "tightening_margin": 0.1829932861142439,
        "predicted_safe_fraction": 0.9875,
        "observed_safe_fraction": 0.8975
      }
    ],
    "residual_margin_correlation": 0.9318449762336007,
    "prediction": "higher trained-model residual produces higher tightening margin",
    "confirmed": true
  }
}