1{
  2  "bench_version": 1,
  3  "track": "tabular",
  4  "model": "mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.006,
 10      "weight_decay": 0.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 15.53538990020752
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "weight_decay": 0.0
 24        },
 25        "mean": 9.415347337722778
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.006,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 7.360140562057495
 33      }
 34    ],
 35    "full": {
 36      "mean": 7.4583141803741455,
 37      "std": 0.6102747432817298,
 38      "per_seed": [
 39        7.289438247680664,
 40        8.00219440460205,
 41        7.081491947174072,
 42        7.067437648773193,
 43        6.286782264709473,
 44        7.646380424499512,
 45        8.093415260314941,
 46        8.199373245239258
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "mean": 7.388119101524353,
 53    "std": 0.6041003222700619,
 54    "per_seed": [
 55      6.846275329589844,
 56      7.8614349365234375,
 57      7.108870506286621,
 58      7.15531587600708,
 59      6.309638023376465,
 60      7.587562084197998,
 61      8.090027809143066,
 62      8.145828247070312
 63    ],
 64    "n": 8
 65  },
 66  "comparison": {
 67    "delta_mean": -0.07019507884979248,
 68    "idea_wins": 5,
 69    "n_pairs": 8,
 70    "per_seed_diffs": [
 71      -0.4431629180908203,
 72      -0.14075946807861328,
 73      0.027378559112548828,
 74      0.08787822723388672,
 75      0.022855758666992188,
 76      -0.05881834030151367,
 77      -0.003387451171875,
 78      -0.05354499816894531
 79    ],
 80    "p_value": 0.2587,
 81    "mde": 0.13852366888165354,
 82    "mde_rel_pct": 1.8573053579073617,
 83    "verdict": "no measurable effect",
 84    "system_worked": false
 85  },
 86  "idea_trials": [
 87    {
 88      "cfg": {
 89        "lr": 0.001,
 90        "weight_decay": 0.0
 91      },
 92      "result": {
 93        "mean": 16.191248655319214,
 94        "std": 1.334610265709453,
 95        "per_seed": [
 96          15.566378593444824,
 97          16.270675659179688,
 98          17.070934295654297,
 99          14.182695388793945,
100          14.616189002990723,
101          15.975017547607422,
102          17.418180465698242,
103          18.42991828918457
104        ],
105        "n": 8
106      }
107    },
108    {
109      "cfg": {
110        "lr": 0.003,
111        "weight_decay": 0.0
112      },
113      "result": {
114        "mean": 9.826780676841736,
115        "std": 1.0954169590780658,
116        "per_seed": [
117          8.976614952087402,
118          11.023777961730957,
119          9.17143726348877,
120          9.166444778442383,
121          8.010053634643555,
122          10.084756851196289,
123          10.998445510864258,
124          11.182714462280273
125        ],
126        "n": 8
127      }
128    },
129    {
130      "cfg": {
131        "lr": 0.006,
132        "weight_decay": 0.0
133      },
134      "result": {
135        "mean": 7.388119101524353,
136        "std": 0.6041003222700619,
137        "per_seed": [
138          6.846275329589844,
139          7.8614349365234375,
140          7.108870506286621,
141          7.15531587600708,
142          6.309638023376465,
143          7.587562084197998,
144          8.090027809143066,
145          8.145828247070312
146        ],
147        "n": 8
148      }
149    }
150  ],
151  "task_match": "optimizer intervention on Friedman regression",
152  "mechanism_signature": {
153    "raw_grad_lag1_corr": 0.3716290132687427,
154    "corrected_grad_lag1_corr": 0.12645231312280522,
155    "baseline_raw_grad_lag1_corr": 0.5400456543155014,
156    "confirmed": true
157  }
158}