1{
  2  "bench_version": 1,
  3  "track": "sequence",
  4  "model": "transformer_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.006
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.2674381099641323
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.2363879233598709
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.006
 27        },
 28        "mean": 0.21260840073227882
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.24213635176420212,
 33      "std": 0.041836180617489935,
 34      "per_seed": [
 35        0.2506037950515747,
 36        0.19614821672439575,
 37        0.1921737939119339,
 38        0.21150779724121094,
 39        0.22115366160869598,
 40        0.2778076231479645,
 41        0.26815494894981384,
 42        0.31954097747802734
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.9611421003937721,
 49    "std": 0.2093114434157431,
 50    "per_seed": [
 51      1.337759017944336,
 52      0.6684795022010803,
 53      0.7549518346786499,
 54      1.1353486776351929,
 55      0.9210895895957947,
 56      0.7781423926353455,
 57      1.0609768629074097,
 58      1.0323889255523682
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.71900574862957,
 64    "idea_wins": 0,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      1.0871552228927612,
 68      0.47233128547668457,
 69      0.562778040766716,
 70      0.9238408803939819,
 71      0.6999359279870987,
 72      0.500334769487381,
 73      0.7928219139575958,
 74      0.7128479480743408
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 0.17777273392901788,
 78    "mde_rel_pct": 73.41844073959494,
 79    "verdict": "idea worse (significant)",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "track_match": "multi-token sequence correlations",
 84    "signature": {
 85      "prediction": "conditioning should narrow q spread (lower beta) and raise lower distortion",
 86      "baseline_seed0": {
 87        "L_q05": 1.2279397010803224,
 88        "U_q95": 15.299684000015256,
 89        "beta": 12.45963786866312,
 90        "mean_q": 6.053372383117676
 91      },
 92      "idea_seed0": {
 93        "L_q05": 0.49697071909904483,
 94        "U_q95": 0.7644718140363693,
 95        "beta": 1.5382632912906329,
 96        "mean_q": 0.6344384551048279
 97      },
 98      "predicted_beta_direction": true,
 99      "predicted_L_direction": false,
100      "confirmed": false
101    },
102    "hyperparameters": {
103      "epochs": 12,
104      "m": 16,
105      "d": 32,
106      "lambda": 0.03,
107      "idea_lr": 0.006
108    }
109  },
110  "idea_sweep": [
111    {
112      "lr": 0.006,
113      "mean_4seed": 0.9741347581148148
114    },
115    {
116      "lr": 0.001,
117      "mean_4seed": 1.035392940044403
118    },
119    {
120      "lr": 0.003,
121      "mean_4seed": 0.9959827661514282
122    }
123  ]
124}