{
  "bench_version": 1,
  "track": "sequence",
  "model": "transformer_tiny",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.006
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.2674381099641323
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.2363879233598709
      },
      {
        "cfg": {
          "lr": 0.006
        },
        "mean": 0.21260840073227882
      }
    ],
    "full": {
      "mean": 0.24213635176420212,
      "std": 0.041836180617489935,
      "per_seed": [
        0.2506037950515747,
        0.19614821672439575,
        0.1921737939119339,
        0.21150779724121094,
        0.22115366160869598,
        0.2778076231479645,
        0.26815494894981384,
        0.31954097747802734
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.9611421003937721,
    "std": 0.2093114434157431,
    "per_seed": [
      1.337759017944336,
      0.6684795022010803,
      0.7549518346786499,
      1.1353486776351929,
      0.9210895895957947,
      0.7781423926353455,
      1.0609768629074097,
      1.0323889255523682
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.71900574862957,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      1.0871552228927612,
      0.47233128547668457,
      0.562778040766716,
      0.9238408803939819,
      0.6999359279870987,
      0.500334769487381,
      0.7928219139575958,
      0.7128479480743408
    ],
    "p_value": 0.0081,
    "mde": 0.17777273392901788,
    "mde_rel_pct": 73.41844073959494,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "mechanism_signature": {
    "track_match": "multi-token sequence correlations",
    "signature": {
      "prediction": "conditioning should narrow q spread (lower beta) and raise lower distortion",
      "baseline_seed0": {
        "L_q05": 1.2279397010803224,
        "U_q95": 15.299684000015256,
        "beta": 12.45963786866312,
        "mean_q": 6.053372383117676
      },
      "idea_seed0": {
        "L_q05": 0.49697071909904483,
        "U_q95": 0.7644718140363693,
        "beta": 1.5382632912906329,
        "mean_q": 0.6344384551048279
      },
      "predicted_beta_direction": true,
      "predicted_L_direction": false,
      "confirmed": false
    },
    "hyperparameters": {
      "epochs": 12,
      "m": 16,
      "d": 32,
      "lambda": 0.03,
      "idea_lr": 0.006
    }
  },
  "idea_sweep": [
    {
      "lr": 0.006,
      "mean_4seed": 0.9741347581148148
    },
    {
      "lr": 0.001,
      "mean_4seed": 1.035392940044403
    },
    {
      "lr": 0.003,
      "mean_4seed": 0.9959827661514282
    }
  ]
}