{
  "bench_version": 1,
  "track": "tabular",
  "model": "mlp_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.006,
      "weight_decay": 0.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "weight_decay": 0.0
        },
        "mean": 25.60164165496826
      },
      {
        "cfg": {
          "lr": 0.003,
          "weight_decay": 0.0
        },
        "mean": 13.860755920410156
      },
      {
        "cfg": {
          "lr": 0.006,
          "weight_decay": 0.0
        },
        "mean": 10.442761182785034
      }
    ],
    "full": {
      "mean": 10.930935978889465,
      "std": 1.3983305692789296,
      "per_seed": [
        10.067484855651855,
        12.066851615905762,
        9.622394561767578,
        10.014313697814941,
        8.787652015686035,
        11.810267448425293,
        12.032318115234375,
        13.046205520629883
      ],
      "n": 8
    }
  },
  "idea": {
    "cfg": {
      "lr": 0.006,
      "weight_decay": 0.0
    },
    "mean": 10.922250509262085,
    "std": 1.396535179145413,
    "per_seed": [
      10.052755355834961,
      12.058212280273438,
      9.612079620361328,
      10.001825332641602,
      8.791247367858887,
      11.804107666015625,
      12.02346420288086,
      13.03431224822998
    ],
    "n": 8,
    "behavior_all_seeds": [
      {
        "raw_spike": 97,
        "clipped_entries": 462,
        "entries": 354888,
        "tail_sum": 0.4687500017462298,
        "k_sum": 4870.490382072394,
        "tau_sum": 3035.111967606231,
        "steps": 72,
        "mean_tail_above_tau": 0.0010850694484866431,
        "mean_kurtosis_bound": 11.274283291834244,
        "mean_tau": 7.025722147236645,
        "clip_fraction": 0.0013018191654831947
      },
      {
        "raw_spike": 102,
        "clipped_entries": 623,
        "entries": 354888,
        "tail_sum": 0.6460449252044782,
        "k_sum": 4870.597906613239,
        "tau_sum": 3113.7841693589585,
        "steps": 72,
        "mean_tail_above_tau": 0.001495474363899255,
        "mean_kurtosis_bound": 11.274532191234348,
        "mean_tau": 7.207833725367959,
        "clip_fraction": 0.0017554834201212778
      },
      {
        "raw_spike": 90,
        "clipped_entries": 539,
        "entries": 354888,
        "tail_sum": 0.4866699246922508,
        "k_sum": 5089.335932393671,
        "tau_sum": 2659.756519132028,
        "steps": 72,
        "mean_tail_above_tau": 0.0011265507516024325,
        "mean_kurtosis_bound": 11.78087021387424,
        "mean_tau": 6.156843794287102,
        "clip_fraction": 0.0015187890263970605
      },
      {
        "raw_spike": 100,
        "clipped_entries": 822,
        "entries": 354888,
        "tail_sum": 0.7038574230391532,
        "k_sum": 4833.979456970323,
        "tau_sum": 2974.376847470511,
        "steps": 72,
        "mean_tail_above_tau": 0.0016292995903684103,
        "mean_kurtosis_bound": 11.189767261505377,
        "mean_tau": 6.885131591366924,
        "clip_fraction": 0.0023162237100155542
      },
      {
        "raw_spike": 81,
        "clipped_entries": 358,
        "entries": 354888,
        "tail_sum": 0.29570312541909516,
        "k_sum": 5140.149971492376,
        "tau_sum": 3046.6057521423736,
        "steps": 72,
        "mean_tail_above_tau": 0.0006844979755071647,
        "mean_kurtosis_bound": 11.8984953043805,
        "mean_tau": 7.052328129959198,
        "clip_fraction": 0.001008768963729402
      },
      {
        "raw_spike": 106,
        "clipped_entries": 853,
        "entries": 354888,
        "tail_sum": 0.7641601584618911,
        "k_sum": 5438.268085443231,
        "tau_sum": 3097.6346601389573,
        "steps": 72,
        "mean_tail_above_tau": 0.001768889255698822,
        "mean_kurtosis_bound": 12.58858353111859,
        "mean_tau": 7.170450602173513,
        "clip_fraction": 0.0024035752124613964
      },
      {
        "raw_spike": 94,
        "clipped_entries": 469,
        "entries": 354888,
        "tail_sum": 0.5587890637107193,
        "k_sum": 5913.583425274048,
        "tau_sum": 3219.2745450983753,
        "steps": 72,
        "mean_tail_above_tau": 0.0012934932030340726,
        "mean_kurtosis_bound": 13.688850521467703,
        "mean_tau": 7.452024409949943,
        "clip_fraction": 0.0013215436982935461
      },
      {
        "raw_spike": 97,
        "clipped_entries": 813,
        "entries": 354888,
        "tail_sum": 0.8062500024680048,
        "k_sum": 5628.396702211232,
        "tau_sum": 3650.706699383914,
        "steps": 72,
        "mean_tail_above_tau": 0.0018663194501574185,
        "mean_kurtosis_bound": 13.028696069933407,
        "mean_tau": 8.45070995227758,
        "clip_fraction": 0.002290863596402245
      }
    ]
  },
  "comparison": {
    "delta_mean": -0.008685469627380371,
    "idea_wins": 7,
    "n_pairs": 8,
    "per_seed_diffs": [
      -0.014729499816894531,
      -0.008639335632324219,
      -0.01031494140625,
      -0.012488365173339844,
      0.0035953521728515625,
      -0.006159782409667969,
      -0.008853912353515625,
      -0.011893272399902344
    ],
    "p_value": 0.01475,
    "mde": 0.004699979815996483,
    "mde_rel_pct": 0.04299704824063914,
    "verdict": "idea better (significant)",
    "system_worked": true
  },
  "math_check": {
    "max_inverse_error": 8.881784197001252e-16,
    "max_transition_polynomial_error": 1.4210854715202004e-14,
    "min_projected_tail_margin": 0.0,
    "passed": true
  },
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.001,
        "weight_decay": 0.0
      },
      "mean": 25.558531761169434
    },
    {
      "cfg": {
        "lr": 0.003,
        "weight_decay": 0.0
      },
      "mean": 13.852189540863037
    },
    {
      "cfg": {
        "lr": 0.006,
        "weight_decay": 0.0
      },
      "mean": 10.431218147277832
    }
  ],
  "protocol_note": "8 paired seeds; baseline sweep and idea sweep share the full learning-rate union; Adam and architecture are otherwise identical.",
  "mechanism_signature": {
    "prediction": "trained gradient-coordinate tail above calibrated tau is controlled near target delta=.02",
    "target_delta": 0.02,
    "observed_mean_tail_above_tau": 0.0013686992548442773,
    "mean_clip_fraction": 0.0017396333491129594,
    "mean_kurtosis_bound": 12.09050979816855,
    "confirmed": true
  }
}