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": 0.0001,
 11      "epochs": 12
 12    },
 13    "sweep": [
 14      {
 15        "cfg": {
 16          "lr": 0.0015,
 17          "weight_decay": 0.0,
 18          "epochs": 12
 19        },
 20        "mean": 13.675705671310425
 21      },
 22      {
 23        "cfg": {
 24          "lr": 0.003,
 25          "weight_decay": 0.0,
 26          "epochs": 12
 27        },
 28        "mean": 9.352656126022339
 29      },
 30      {
 31        "cfg": {
 32          "lr": 0.006,
 33          "weight_decay": 0.0,
 34          "epochs": 12
 35        },
 36        "mean": 7.226524114608765
 37      },
 38      {
 39        "cfg": {
 40          "lr": 0.0015,
 41          "weight_decay": 0.0001,
 42          "epochs": 12
 43        },
 44        "mean": 13.609990000724792
 45      },
 46      {
 47        "cfg": {
 48          "lr": 0.003,
 49          "weight_decay": 0.0001,
 50          "epochs": 12
 51        },
 52        "mean": 9.076743841171265
 53      },
 54      {
 55        "cfg": {
 56          "lr": 0.006,
 57          "weight_decay": 0.0001,
 58          "epochs": 12
 59        },
 60        "mean": 7.0532397627830505
 61      }
 62    ],
 63    "full": {
 64      "mean": 7.0532397627830505,
 65      "std": 0.2264412705403704,
 66      "per_seed": [
 67        6.572167873382568,
 68        7.1756792068481445,
 69        7.042057037353516,
 70        7.032951831817627,
 71        7.018033504486084,
 72        7.022153854370117,
 73        7.112776756286621,
 74        7.450098037719727
 75      ],
 76      "n": 8
 77    }
 78  },
 79  "idea": {
 80    "mean": 7.014131605625153,
 81    "std": 0.24308566746895208,
 82    "per_seed": [
 83      6.6641693115234375,
 84      7.260889053344727,
 85      6.737545013427734,
 86      6.856451034545898,
 87      6.927297592163086,
 88      7.039077281951904,
 89      7.303320407867432,
 90      7.324303150177002
 91    ],
 92    "n": 8
 93  },
 94  "comparison": {
 95    "delta_mean": -0.03910815715789795,
 96    "idea_wins": 4,
 97    "n_pairs": 8,
 98    "per_seed_diffs": [
 99      0.09200143814086914,
100      0.08520984649658203,
101      -0.30451202392578125,
102      -0.17650079727172852,
103      -0.09073591232299805,
104      0.01692342758178711,
105      0.19054365158081055,
106      -0.1257948875427246
107    ],
108    "p_value": 0.50145,
109    "mde": 0.13703746835599878,
110    "mde_rel_pct": 1.9429010350546605,
111    "verdict": "no measurable effect",
112    "system_worked": false
113  },
114  "selection_details": "Entropy and validation-risk masks selected from train split; final systems use identical mlp_tiny/training.",
115  "mechanism_signature": {
116    "prediction": "kappa>=0.90 routes away from entropy mask and avoids materially worse trained-model MSE",
117    "observed": [
118      {
119        "seed": 0,
120        "kappa": 1.0,
121        "gamma": 13.140283130679085,
122        "used_entropy": false,
123        "predicted_fallback": true,
124        "observed_subset_minus_full_mse": 0.09200143814086914
125      },
126      {
127        "seed": 1,
128        "kappa": 1.0,
129        "gamma": 16.30871552694337,
130        "used_entropy": false,
131        "predicted_fallback": true,
132        "observed_subset_minus_full_mse": 0.08520984649658203
133      },
134      {
135        "seed": 2,
136        "kappa": 1.0,
137        "gamma": 12.264759413539208,
138        "used_entropy": false,
139        "predicted_fallback": true,
140        "observed_subset_minus_full_mse": -0.30451202392578125
141      },
142      {
143        "seed": 3,
144        "kappa": 1.0,
145        "gamma": 13.648493168331017,
146        "used_entropy": false,
147        "predicted_fallback": true,
148        "observed_subset_minus_full_mse": -0.17650079727172852
149      },
150      {
151        "seed": 4,
152        "kappa": 1.0,
153        "gamma": 15.0961523290884,
154        "used_entropy": false,
155        "predicted_fallback": true,
156        "observed_subset_minus_full_mse": -0.09073591232299805
157      },
158      {
159        "seed": 5,
160        "kappa": 1.0,
161        "gamma": 11.799654276914342,
162        "used_entropy": false,
163        "predicted_fallback": true,
164        "observed_subset_minus_full_mse": 0.01692342758178711
165      },
166      {
167        "seed": 6,
168        "kappa": 1.0,
169        "gamma": 15.35830490498792,
170        "used_entropy": false,
171        "predicted_fallback": true,
172        "observed_subset_minus_full_mse": 0.19054365158081055
173      },
174      {
175        "seed": 7,
176        "kappa": 1.0,
177        "gamma": 13.753987138531764,
178        "used_entropy": false,
179        "predicted_fallback": true,
180        "observed_subset_minus_full_mse": -0.1257948875427246
181      }
182    ],
183    "high_kappa_n": 8,
184    "high_kappa_observed_mean_delta": -0.03910815715789795,
185    "confirmed": true
186  }
187}