{
  "bench_version": 1,
  "track": "sequence",
  "model": "transformer_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.0015
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.0015
        },
        "mean": 0.6402924656867981
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.7056357711553574
      },
      {
        "cfg": {
          "lr": 0.006
        },
        "mean": 0.9283334761857986
      }
    ],
    "full": {
      "mean": 0.6229188442230225,
      "std": 0.06931051155059516,
      "per_seed": [
        0.7044264674186707,
        0.5437753200531006,
        0.5856346487998962,
        0.7273334264755249,
        0.5151234269142151,
        0.6567925214767456,
        0.6100528240203857,
        0.6402121186256409
      ],
      "n": 8
    }
  },
  "idea": {
    "best_cfg": {
      "lr": 0.0015,
      "beta": 0.2,
      "eps": 0.15
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.0015,
          "beta": 0.2,
          "eps": 0.15
        },
        "mean": 0.4291221722960472
      },
      {
        "cfg": {
          "lr": 0.003,
          "beta": 0.3,
          "eps": 0.15
        },
        "mean": 0.4798029735684395
      },
      {
        "cfg": {
          "lr": 0.006,
          "beta": 0.4,
          "eps": 0.2
        },
        "mean": 0.6943782418966293
      }
    ],
    "per_seed": [
      0.49007120728492737,
      0.31841030716896057,
      0.4649355113506317,
      0.4430716633796692,
      0.3909056484699249,
      0.4021309018135071,
      0.49423667788505554,
      0.3951165974140167
    ]
  },
  "comparison": {
    "delta_mean": -0.19805902987718582,
    "idea_wins": 8,
    "n_pairs": 8,
    "per_seed_diffs": [
      -0.2143552601337433,
      -0.22536501288414001,
      -0.12069913744926453,
      -0.2842617630958557,
      -0.12421777844429016,
      -0.2546616196632385,
      -0.1158161461353302,
      -0.24509552121162415
    ],
    "p_value": 0.0081,
    "mde": 0.05658672011309537,
    "mde_rel_pct": 9.084123981459733,
    "verdict": "idea better (significant)",
    "system_worked": true
  },
  "mechanism_signature": {
    "prediction": "relational compatibility reduces incompatible-token attention",
    "predicted": {
      "partial_row_upper_bound": 0.03125,
      "beta": 0.2
    },
    "observed": {
      "baseline_test_mse_mean": 0.6229188442230225,
      "idea_test_mse_mean": 0.42485981434583664
    },
    "confirmed": false
  },
  "protocol_notes": {
    "track_reason": "The sequence track is multi-token temporal forecasting with transformer attention, directly matching the proposed mechanism.",
    "reduced_budget": "600 train/300 test and 5 epochs due to PGW O(L^4) cost; paired seeds remain 8; baseline sweep uses the same lr union.",
    "elapsed_sec": 37.595829248428345
  }
}