{
  "bench_version": 1,
  "track": "sequence",
  "model": "transformer_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.001,
      "epochs": 6,
      "weight_decay": 0.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "epochs": 6,
          "weight_decay": 0.0
        },
        "mean": 0.46786613017320633
      },
      {
        "cfg": {
          "lr": 0.003,
          "epochs": 6,
          "weight_decay": 0.0
        },
        "mean": 0.5299046412110329
      },
      {
        "cfg": {
          "lr": 0.006,
          "epochs": 6,
          "weight_decay": 0.0
        },
        "mean": 0.7419369965791702
      }
    ],
    "full": {
      "mean": 0.4766220450401306,
      "std": 0.09934194575564281,
      "per_seed": [
        0.5715442299842834,
        0.39496564865112305,
        0.3930272161960602,
        0.5119274258613586,
        0.4745565354824066,
        0.4187285602092743,
        0.6776894927024841,
        0.37053725123405457
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.5983163416385651,
    "std": 0.16587939812424876,
    "per_seed": [
      0.3822963237762451,
      0.5229498744010925,
      0.8081433773040771,
      0.863442063331604,
      0.6520702838897705,
      0.45361945033073425,
      0.6645830869674683,
      0.4394262731075287
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.12169429659843445,
    "idea_wins": 2,
    "n_pairs": 8,
    "per_seed_diffs": [
      -0.18924790620803833,
      0.12798422574996948,
      0.41511616110801697,
      0.35151463747024536,
      0.1775137484073639,
      0.03489089012145996,
      -0.01310640573501587,
      0.06888902187347412
    ],
    "p_value": 0.1299,
    "mde": 0.16337849985483555,
    "mde_rel_pct": 34.27841862435872,
    "verdict": "no significant win",
    "system_worked": false
  },
  "mechanism_signature": {
    "idea_sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "epochs": 6,
          "weight_decay": 0.0
        },
        "result": {
          "mean": 0.5983163416385651,
          "std": 0.16587939812424876,
          "per_seed": [
            0.3822963237762451,
            0.5229498744010925,
            0.8081433773040771,
            0.863442063331604,
            0.6520702838897705,
            0.45361945033073425,
            0.6645830869674683,
            0.4394262731075287
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.003,
          "epochs": 6,
          "weight_decay": 0.0
        },
        "result": {
          "mean": 2.1142318695783615,
          "std": 0.7959891886633405,
          "per_seed": [
            2.495609998703003,
            2.4600517749786377,
            0.8293719291687012,
            1.1857432126998901,
            1.480914831161499,
            3.2832841873168945,
            2.778350830078125,
            2.4005281925201416
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.006,
          "epochs": 6,
          "weight_decay": 0.0
        },
        "result": {
          "mean": 5.809824541211128,
          "std": 2.711991843326257,
          "per_seed": [
            4.945932388305664,
            11.078203201293945,
            1.813473105430603,
            5.887286186218262,
            2.609509229660034,
            6.428954124450684,
            7.734877109527588,
            5.980360984802246
          ],
          "n": 8
        }
      }
    ],
    "observed_best_cfg": {
      "lr": 0.001,
      "epochs": 6,
      "weight_decay": 0.0
    },
    "math_check": [
      {
        "n": 2,
        "norm_ratio": 1.0000001192092896,
        "inverse_max_error": 2.384185791015625e-07
      },
      {
        "n": 4,
        "norm_ratio": 1.0,
        "inverse_max_error": 4.76837158203125e-07
      },
      {
        "n": 8,
        "norm_ratio": 1.0000001192092896,
        "inverse_max_error": 4.76837158203125e-07
      },
      {
        "n": 16,
        "norm_ratio": 1.0000001192092896,
        "inverse_max_error": 7.152557373046875e-07
      },
      {
        "n": 32,
        "norm_ratio": 1.0,
        "inverse_max_error": 9.5367431640625e-07
      }
    ],
    "mechanism_signature": {
      "prediction": "trained isometric token mixer preserves feature-token Euclidean norm and has exact adjoint inverse",
      "predicted_norm_ratio": 1.0,
      "observed_norm_ratio_seed0": 1.0,
      "observed_inverse_max_error_seed0": 1.1920928955078125e-06,
      "baseline_observed_norm_ratio_seed0": 0.5025865435600281,
      "confirmed": true,
      "trained_model_metric_seed0": 0.3822963237762451
    }
  }
}