Descent-Certified LMO Sign Switching / bench_report.json

Failed on benchmark

Raw ⬇ ZIP
  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      "beta": 0.9
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "beta": 0.8
 17        },
 18        "mean": 137.9729061126709
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "beta": 0.9
 24        },
 25        "mean": 138.0125026702881
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.003,
 30          "beta": 0.8
 31        },
 32        "mean": 45.94260215759277
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "beta": 0.9
 38        },
 39        "mean": 45.709683418273926
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.006,
 44          "beta": 0.8
 45        },
 46        "mean": 20.55884027481079
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.006,
 51          "beta": 0.9
 52        },
 53        "mean": 20.514626026153564
 54      }
 55    ],
 56    "full": {
 57      "mean": 21.575043439865112,
 58      "std": 1.9157591398399918,
 59      "per_seed": [
 60        22.02391815185547,
 61        17.909542083740234,
 62        21.564067840576172,
 63        20.560976028442383,
 64        20.68805694580078,
 65        23.367746353149414,
 66        21.641395568847656,
 67        24.84464454650879
 68      ],
 69      "n": 8
 70    }
 71  },
 72  "idea": {
 73    "mean": 21.585941553115845,
 74    "std": 1.8841984845237936,
 75    "per_seed": [
 76      22.23833465576172,
 77      17.9639835357666,
 78      21.634672164916992,
 79      20.441940307617188,
 80      20.853069305419922,
 81      23.09563446044922,
 82      21.598304748535156,
 83      24.86159324645996
 84    ],
 85    "n": 8
 86  },
 87  "comparison": {
 88    "delta_mean": 0.010898113250732422,
 89    "idea_wins": 3,
 90    "n_pairs": 8,
 91    "per_seed_diffs": [
 92      0.21441650390625,
 93      0.05444145202636719,
 94      0.07060432434082031,
 95      -0.11903572082519531,
 96      0.16501235961914062,
 97      -0.2721118927001953,
 98      -0.0430908203125,
 99      0.016948699951171875
100    ],
101    "p_value": 0.8426,
102    "mde": 0.13051985349480585,
103    "mde_rel_pct": 0.6049575467071313,
104    "verdict": "no measurable effect",
105    "system_worked": false
106  },
107  "mechanism_signature": {
108    "prediction": "heldout alignment gating routes post when rho>=tau and reduces negative heldout alignment",
109    "observed": {
110      "post_fraction": 0.9606481481481481,
111      "rho": 0.2859362520431751,
112      "post_negative": 0.039351851851851846,
113      "chosen_negative": 0.03306878306878307,
114      "predicted_fraction": 0.9606481481481481,
115      "confirmed": true
116    },
117    "tau": 0.0,
118    "trained_models": true
119  },
120  "idea_sweep": [
121    {
122      "cfg": {
123        "lr": 0.001,
124        "beta": 0.9,
125        "tau": 0.0
126      },
127      "mean": 136.1921682357788,
128      "per_seed": [
129        149.82797241210938,
130        138.97898864746094,
131        137.3268585205078,
132        125.91619110107422,
133        135.9979705810547,
134        124.63673400878906,
135        128.59912109375,
136        148.25350952148438
137      ]
138    },
139    {
140      "cfg": {
141        "lr": 0.001,
142        "beta": 0.9,
143        "tau": 0.05
144      },
145      "mean": 136.1921682357788,
146      "per_seed": [
147        149.82797241210938,
148        138.97898864746094,
149        137.3268585205078,
150        125.91619110107422,
151        135.9979705810547,
152        124.63673400878906,
153        128.59912109375,
154        148.25350952148438
155      ]
156    },
157    {
158      "cfg": {
159        "lr": 0.001,
160        "beta": 0.9,
161        "tau": 0.1
162      },
163      "mean": 136.7702980041504,
164      "per_seed": [
165        150.35202026367188,
166        139.48507690429688,
167        137.9378662109375,
168        125.91619110107422,
169        136.5552520751953,
170        125.13619232177734,
171        129.91458129882812,
172        148.86520385742188
173      ]
174    },
175    {
176      "cfg": {
177        "lr": 0.003,
178        "beta": 0.9,
179        "tau": 0.0
180      },
181      "mean": 44.54532766342163,
182      "per_seed": [
183        53.77455520629883,
184        45.56505584716797,
185        42.8125,
186        40.686622619628906,
187        39.3796501159668,
188        41.54080581665039,
189        41.00275421142578,
190        51.600677490234375
191      ]
192    },
193    {
194      "cfg": {
195        "lr": 0.003,
196        "beta": 0.9,
197        "tau": 0.05
198      },
199      "mean": 44.54532766342163,
200      "per_seed": [
201        53.77455520629883,
202        45.56505584716797,
203        42.8125,
204        40.686622619628906,
205        39.3796501159668,
206        41.54080581665039,
207        41.00275421142578,
208        51.600677490234375
209      ]
210    },
211    {
212      "cfg": {
213        "lr": 0.003,
214        "beta": 0.9,
215        "tau": 0.1
216      },
217      "mean": 46.63834524154663,
218      "per_seed": [
219        54.938777923583984,
220        47.869895935058594,
221        44.608856201171875,
222        41.21122360229492,
223        42.689353942871094,
224        43.306400299072266,
225        44.728363037109375,
226        53.75389099121094
227      ]
228    },
229    {
230      "cfg": {
231        "lr": 0.006,
232        "beta": 0.9,
233        "tau": 0.0
234      },
235      "mean": 21.585941553115845,
236      "per_seed": [
237        22.23833465576172,
238        17.9639835357666,
239        21.634672164916992,
240        20.441940307617188,
241        20.853069305419922,
242        23.09563446044922,
243        21.598304748535156,
244        24.86159324645996
245      ]
246    },
247    {
248      "cfg": {
249        "lr": 0.006,
250        "beta": 0.9,
251        "tau": 0.05
252      },
253      "mean": 21.613163232803345,
254      "per_seed": [
255        22.23833465576172,
256        17.98293685913086,
257        21.707622528076172,
258        20.458215713500977,
259        20.899755477905273,
260        23.078847885131836,
261        21.629314422607422,
262        24.9102783203125
263      ]
264    },
265    {
266      "cfg": {
267        "lr": 0.006,
268        "beta": 0.9,
269        "tau": 0.1
270      },
271      "mean": 22.09021544456482,
272      "per_seed": [
273        22.61715316772461,
274        18.391338348388672,
275        21.940710067749023,
276        20.666790008544922,
277        21.886329650878906,
278        23.460872650146484,
279        22.199935913085938,
280        25.55859375
281      ]
282    }
283  ],
284  "custom_track": null
285}