{
  "bench_version": 1,
  "track": "sequence",
  "model": "transformer_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01,
      "temperature": 0.8,
      "hazard_bias": 0.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "temperature": 0.8,
          "hazard_bias": 0.0
        },
        "mean": 1.0667133182287216
      },
      {
        "cfg": {
          "lr": 0.001,
          "temperature": 1.2,
          "hazard_bias": 0.0
        },
        "mean": 1.0630156695842743
      },
      {
        "cfg": {
          "lr": 0.003,
          "temperature": 0.8,
          "hazard_bias": 0.0
        },
        "mean": 1.07551509141922
      },
      {
        "cfg": {
          "lr": 0.003,
          "temperature": 1.2,
          "hazard_bias": 0.0
        },
        "mean": 1.0769807696342468
      },
      {
        "cfg": {
          "lr": 0.01,
          "temperature": 0.8,
          "hazard_bias": 0.0
        },
        "mean": 1.0265950411558151
      },
      {
        "cfg": {
          "lr": 0.01,
          "temperature": 1.2,
          "hazard_bias": 0.0
        },
        "mean": 1.0694856494665146
      }
    ],
    "full": {
      "mean": 1.0413843393325806,
      "std": 0.07579068628025519,
      "per_seed": [
        1.030417799949646,
        0.8935723900794983,
        1.0985037088394165,
        1.0838862657546997,
        0.9647193551063538,
        1.0631779432296753,
        1.1534550189971924,
        1.0433422327041626
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 1.0507365465164185,
    "std": 0.07984042752181207,
    "per_seed": [
      1.0360348224639893,
      0.9041898846626282,
      1.1562472581863403,
      1.0952754020690918,
      0.9761890769004822,
      1.0409249067306519,
      1.1531434059143066,
      1.0438876152038574
    ],
    "n": 8,
    "chosen_cfg": {
      "lr": 0.01,
      "temperature": 0.8,
      "hazard_bias": 0.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.01,
          "temperature": 0.8,
          "hazard_bias": -1.0
        },
        "result": {
          "mean": 1.0596621185541153,
          "std": 0.07361314701141786,
          "per_seed": [
            1.071687936782837,
            0.9170374870300293,
            1.136078119277954,
            1.1125483512878418,
            0.9964454174041748,
            1.02909517288208,
            1.15607750415802,
            1.0583269596099854
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.01,
          "temperature": 0.8,
          "hazard_bias": 0.0
        },
        "result": {
          "mean": 1.0507365465164185,
          "std": 0.07984042752181207,
          "per_seed": [
            1.0360348224639893,
            0.9041898846626282,
            1.1562472581863403,
            1.0952754020690918,
            0.9761890769004822,
            1.0409249067306519,
            1.1531434059143066,
            1.0438876152038574
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.01,
          "temperature": 0.8,
          "hazard_bias": 1.0
        },
        "result": {
          "mean": 1.0628163740038872,
          "std": 0.06874408412093434,
          "per_seed": [
            1.0399799346923828,
            0.9146118760108948,
            1.1155239343643188,
            1.1016066074371338,
            1.0239859819412231,
            1.0964356660842896,
            1.1542212963104248,
            1.0561656951904297
          ],
          "n": 8
        }
      }
    ]
  },
  "comparison": {
    "delta_mean": 0.00935220718383789,
    "idea_wins": 2,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.005617022514343262,
      0.010617494583129883,
      0.05774354934692383,
      0.01138913631439209,
      0.011469721794128418,
      -0.022253036499023438,
      -0.0003116130828857422,
      0.0005453824996948242
    ],
    "p_value": 0.3073,
    "mde": 0.018792208195343812,
    "mde_rel_pct": 1.8045410791742527,
    "verdict": "no measurable effect",
    "system_worked": false
  },
  "track_justification": "Sequence forecasting has multi-token correlations and regime persistence, matching semi-Markov routing.",
  "mechanism_signature": {
    "predicted_mean_hazard": 0.5708998311311007,
    "observed_trace_change": 0.009716346365166828,
    "predicted_vs_observed_ratio": 58.75663646345304,
    "age_mean": 0.9239334501326084,
    "per_seed": [
      {
        "predicted_mean_hazard": 0.7784030437469482,
        "observed_trace_change": 0.016614258289337158,
        "age_mean": 0.2825595736503601
      },
      {
        "predicted_mean_hazard": 0.6146835684776306,
        "observed_trace_change": 0.0018246846739202738,
        "age_mean": 0.6145921945571899
      },
      {
        "predicted_mean_hazard": 0.622718870639801,
        "observed_trace_change": 0.004094474017620087,
        "age_mean": 0.594530463218689
      },
      {
        "predicted_mean_hazard": 0.5631765127182007,
        "observed_trace_change": 0.00499045429751277,
        "age_mean": 0.7603605389595032
      },
      {
        "predicted_mean_hazard": 0.5672649145126343,
        "observed_trace_change": 0.012080281972885132,
        "age_mean": 0.7463456392288208
      },
      {
        "predicted_mean_hazard": 0.5145835280418396,
        "observed_trace_change": 0.006644372828304768,
        "age_mean": 0.9115980863571167
      },
      {
        "predicted_mean_hazard": 0.23928262293338776,
        "observed_trace_change": 0.017612455412745476,
        "age_mean": 2.9931130409240723
      },
      {
        "predicted_mean_hazard": 0.667085587978363,
        "observed_trace_change": 0.01386978942900896,
        "age_mean": 0.48836806416511536
      }
    ],
    "confirmed": false
  }
}