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": 0.002,
 10      "weight_decay": 0.0001
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 140.17193984985352
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "weight_decay": 0.0001
 24        },
 25        "mean": 140.17272758483887
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.002,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 46.896180152893066
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.002,
 37          "weight_decay": 0.0001
 38        },
 39        "mean": 46.89336395263672
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.003,
 44          "weight_decay": 0.0
 45        },
 46        "mean": 425.859375
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.003,
 51          "weight_decay": 0.0001
 52        },
 53        "mean": 425.85955810546875
 54      },
 55      {
 56        "cfg": {
 57          "lr": 0.006,
 58          "weight_decay": 0.0
 59        },
 60        "mean": 150.0003662109375
 61      },
 62      {
 63        "cfg": {
 64          "lr": 0.006,
 65          "weight_decay": 0.0001
 66        },
 67        "mean": 150.00535202026367
 68      }
 69    ],
 70    "full": {
 71      "mean": 53.96195363998413,
 72      "std": 40.99246488816214,
 73      "per_seed": [
 74        30.952533721923828,
 75        22.93033218383789,
 76        83.2318344116211,
 77        50.45875549316406,
 78        149.6821746826172,
 79        21.348867416381836,
 80        47.70417022705078,
 81        25.386960983276367
 82      ],
 83      "n": 8
 84    }
 85  },
 86  "idea": {
 87    "mean": 45.54406785964966,
 88    "std": 8.338173492397852,
 89    "per_seed": [
 90      43.00838851928711,
 91      44.718257904052734,
 92      28.018310546875,
 93      49.26393508911133,
 94      56.6592903137207,
 95      41.586570739746094,
 96      54.953060150146484,
 97      46.14472961425781
 98    ],
 99    "n": 8,
100    "selected_cfg": {
101      "lr": 0.002,
102      "weight_decay": 0.0001
103    }
104  },
105  "comparison": {
106    "delta_mean": -8.417885780334473,
107    "idea_wins": 3,
108    "n_pairs": 8,
109    "per_seed_diffs": [
110      12.055854797363281,
111      21.787925720214844,
112      -55.213523864746094,
113      -1.1948204040527344,
114      -93.02288436889648,
115      20.237703323364258,
116      7.248889923095703,
117      20.757768630981445
118    ],
119    "p_value": 0.60105,
120    "mde": 35.54445282513561,
121    "mde_rel_pct": 65.86946992741618,
122    "verdict": "no significant win",
123    "system_worked": false
124  },
125  "mechanism_signature": {
126    "math_check": {
127      "bound_max_ratio": 0.827890485216382,
128      "bound": 1.0,
129      "contraction_predicted": [
130        0.9619047619047619,
131        0.8666666666666667,
132        0.6,
133        0.36
134      ],
135      "contraction_measured": [
136        0.9619047619047619,
137        0.8666666666666666,
138        0.5999999999999999,
139        0.35999999999999976
140      ],
141      "max_contraction_error": 2.220446049250313e-16
142    },
143    "trained_model_signature": {
144      "mean_observed_diag_gramian": 0.2610868251649663,
145      "mean_update_factor_predicted": 0.8549541880687078,
146      "mean_update_factor_observed": 0.8544371028741201,
147      "relative_factor_error": 0.0006048104118371221,
148      "confirmed": true
149    },
150    "idea_configs": [
151      {
152        "cfg": {
153          "lr": 0.002,
154          "weight_decay": 0.0001
155        },
156        "result": {
157          "mean": 45.54406785964966,
158          "std": 8.338173492397852,
159          "per_seed": [
160            43.00838851928711,
161            44.718257904052734,
162            28.018310546875,
163            49.26393508911133,
164            56.6592903137207,
165            41.586570739746094,
166            54.953060150146484,
167            46.14472961425781
168          ],
169          "n": 8,
170          "selected_cfg": {
171            "lr": 0.002,
172            "weight_decay": 0.0001
173          }
174        }
175      },
176      {
177        "cfg": {
178          "lr": 0.002,
179          "weight_decay": 0.0001
180        },
181        "result": {
182          "mean": 45.54406785964966,
183          "std": 8.338173492397852,
184          "per_seed": [
185            43.00838851928711,
186            44.718257904052734,
187            28.018310546875,
188            49.26393508911133,
189            56.6592903137207,
190            41.586570739746094,
191            54.953060150146484,
192            46.14472961425781
193          ],
194          "n": 8
195        }
196      },
197      {
198        "cfg": {
199          "lr": 0.006,
200          "weight_decay": 0.0001
201        },
202        "result": {
203          "mean": 53.81347846984863,
204          "std": 9.649444093972434,
205          "per_seed": [
206            54.86537170410156,
207            67.9201431274414,
208            36.08623123168945,
209            58.39324188232422,
210            65.55195617675781,
211            47.311336517333984,
212            51.460205078125,
213            48.919342041015625
214          ],
215          "n": 8
216        }
217      }
218    ]
219  },
220  "protocol_notes": "Optimizer modification on structurally matched Friedman tabular regression; paired seeds 0-7, batch 128, 12 epochs, same architecture and lr/weight-decay union."
221}