{
  "bench_version": 1,
  "track": "sequence",
  "model": "transformer_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.001,
      "epochs": 8,
      "rho": 0.0,
      "beta": 0.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "epochs": 8,
          "rho": 0.0,
          "beta": 0.0
        },
        "mean": 0.5321156531572342
      },
      {
        "cfg": {
          "lr": 0.003,
          "epochs": 8,
          "rho": 0.0,
          "beta": 0.0
        },
        "mean": 0.6582250595092773
      },
      {
        "cfg": {
          "lr": 0.006,
          "epochs": 8,
          "rho": 0.0,
          "beta": 0.0
        },
        "mean": 0.91780074685812
      }
    ],
    "full": {
      "mean": 0.5321156531572342,
      "std": 0.03988514268446927,
      "per_seed": [
        0.5299221873283386,
        0.45885390043258667,
        0.5064432621002197,
        0.5547861456871033,
        0.5245285630226135,
        0.5539869070053101,
        0.6065487861633301,
        0.5218554735183716
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.5394741781055927,
    "std": 0.0655720338898959,
    "per_seed": [
      0.6348236799240112,
      0.39886191487312317,
      0.5056540369987488,
      0.5536980628967285,
      0.5338740348815918,
      0.5505685210227966,
      0.602857768535614,
      0.5354554057121277
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.007358524948358536,
    "idea_wins": 5,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.10490149259567261,
      -0.0599919855594635,
      -0.0007892251014709473,
      -0.0010880827903747559,
      0.009345471858978271,
      -0.0034183859825134277,
      -0.0036910176277160645,
      0.013599932193756104
    ],
    "p_value": 0.6281,
    "mde": 0.03801589735734751,
    "mde_rel_pct": 7.1442922477069555,
    "verdict": "no significant win",
    "system_worked": false
  },
  "mechanism_signature": {
    "track_rationale": "Sequence forecast contains multi-token temporal correlations and matches the proposed per-token fading-memory transformer gate.",
    "observed_best_cfg": {
      "lr": 0.001,
      "epochs": 8,
      "rho": 0.85,
      "beta": 0.5
    },
    "idea_sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "epochs": 8,
          "rho": 0.85,
          "beta": 0.5
        },
        "result": {
          "mean": 0.5394741781055927,
          "std": 0.0655720338898959,
          "per_seed": [
            0.6348236799240112,
            0.39886191487312317,
            0.5056540369987488,
            0.5536980628967285,
            0.5338740348815918,
            0.5505685210227966,
            0.602857768535614,
            0.5354554057121277
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.003,
          "epochs": 8,
          "rho": 0.9,
          "beta": 0.5
        },
        "result": {
          "mean": 0.6679390706121922,
          "std": 0.11291799270623928,
          "per_seed": [
            0.57618248462677,
            0.4957140386104584,
            0.6142106652259827,
            0.6856930255889893,
            0.6076048016548157,
            0.8247736692428589,
            0.6928988695144653,
            0.846435010433197
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.006,
          "epochs": 8,
          "rho": 0.9,
          "beta": 1.0
        },
        "result": {
          "mean": 0.9237265363335609,
          "std": 0.17665093311633415,
          "per_seed": [
            1.304537057876587,
            0.6689846515655518,
            0.846419632434845,
            1.0522245168685913,
            0.8211356401443481,
            0.86773282289505,
            0.9556908011436462,
            0.8730871677398682
          ],
          "n": 8
        }
      }
    ],
    "mechanism_signature": {
      "trained_model_seed": 0,
      "metric": 0.6348236799240112,
      "initial_gain": 0.9424585700035095,
      "final_gain": 0.7104766368865967,
      "mean_state": 0.8186129927635193,
      "predicted_steady_gain": 0.7095688571417152,
      "fitted_state_pole": 1.0177967548370361,
      "rho": 0.85,
      "beta": 0.5,
      "confirmed": false
    },
    "math_check": {
      "rows": [
        {
          "rho": 0.85,
          "steady_gain_abs_error": 3.3306690738754696e-16,
          "half_observed": 5,
          "half_predicted": 5
        },
        {
          "rho": 0.9,
          "steady_gain_abs_error": 4.773959005888173e-15,
          "half_observed": 7,
          "half_predicted": 7
        },
        {
          "rho": 0.9,
          "steady_gain_abs_error": 4.385380947269368e-15,
          "half_observed": 7,
          "half_predicted": 7
        }
      ],
      "confirmed": true
    },
    "protocol": {
      "seeds": [
        0,
        1,
        2,
        3,
        4,
        5,
        6,
        7
      ],
      "epochs": 8,
      "baseline_lr_grid": [
        0.001,
        0.003,
        0.006
      ],
      "idea_lr_grid": [
        0.001,
        0.003,
        0.006
      ]
    }
  }
}