Capitalization-Efficiency Monitor / bench_report.json

Mechanism confirmed, baseline not beaten

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    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.0015
 15        },
 16        "mean": 8.577727317810059
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 6.8584965467453
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.006
 27        },
 28        "mean": 6.626367449760437
 29      }
 30    ],
 31    "full": {
 32      "mean": 6.720197975635529,
 33      "std": 0.3792385031605589,
 34      "per_seed": [
 35        6.333833694458008,
 36        7.1719770431518555,
 37        6.258625507354736,
 38        6.741033554077148,
 39        6.219150543212891,
 40        6.970664978027344,
 41        6.840729713439941,
 42        7.225568771362305
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "best_cfg": {
 49      "lr": 0.006,
 50      "threshold": 0.02
 51    },
 52    "sweep": [
 53      {
 54        "cfg": {
 55          "lr": 0.0015,
 56          "threshold": 0.02
 57        },
 58        "mean": 8.962467908859253,
 59        "per_seed": [
 60          7.58711051940918,
 61          9.554247856140137,
 62          8.6873140335083,
 63          8.482236862182617,
 64          8.62455940246582,
 65          8.597721099853516,
 66          10.342260360717773,
 67          9.82429313659668
 68        ]
 69      },
 70      {
 71        "cfg": {
 72          "lr": 0.003,
 73          "threshold": 0.02
 74        },
 75        "mean": 7.0224193930625916,
 76        "per_seed": [
 77          6.4233717918396,
 78          7.385267734527588,
 79          6.821650505065918,
 80          6.856407165527344,
 81          7.057806015014648,
 82          6.954143047332764,
 83          7.078849792480469,
 84          7.601859092712402
 85        ]
 86      },
 87      {
 88        "cfg": {
 89          "lr": 0.006,
 90          "threshold": 0.02
 91        },
 92        "mean": 6.847910106182098,
 93        "per_seed": [
 94          6.578519821166992,
 95          7.120673179626465,
 96          6.699369430541992,
 97          6.741033554077148,
 98          6.219150543212891,
 99          6.990142345428467,
100          6.840729713439941,
101          7.593662261962891
102        ]
103      }
104    ],
105    "per_seed": [
106      6.578519821166992,
107      7.120673179626465,
108      6.699369430541992,
109      6.741033554077148,
110      6.219150543212891,
111      6.990142345428467,
112      6.840729713439941,
113      7.593662261962891
114    ],
115    "mean": 6.847910106182098,
116    "std": 0.3795102408701731,
117    "n": 8
118  },
119  "comparison": {
120    "delta_mean": 0.12771213054656982,
121    "idea_wins": 1,
122    "n_pairs": 8,
123    "per_seed_diffs": [
124      0.24468612670898438,
125      -0.051303863525390625,
126      0.44074392318725586,
127      0.0,
128      0.0,
129      0.019477367401123047,
130      0.0,
131      0.36809349060058594
132    ],
133    "p_value": 0.1856,
134    "mde": 0.16181177964695484,
135    "mde_rel_pct": 2.407842450975595,
136    "verdict": "no significant win",
137    "system_worked": false
138  },
139  "mechanism_signature": {
140    "quantity": "trained-model epoch displacement KL and held-out validation value/ledger",
141    "prediction": "late low-value updates acquire positive KL and have eta near zero or negative",
142    "observed_mean_info": 35.882227032415834,
143    "observed_mean_late_eta": -0.5215707736578389,
144    "observed_rollbacks": 1.25,
145    "confirmed": true,
146    "per_seed": [
147      {
148        "seed": 0,
149        "idea_info": 40.445681335271644,
150        "idea_late_eta": -0.8021093204912924,
151        "idea_rollbacks": 2
152      },
153      {
154        "seed": 1,
155        "idea_info": 30.49450051147865,
156        "idea_late_eta": -5.761187156173132,
157        "idea_rollbacks": 2
158      },
159      {
160        "seed": 2,
161        "idea_info": 37.59771323140545,
162        "idea_late_eta": 0.012780262934585526,
163        "idea_rollbacks": 2
164      },
165      {
166        "seed": 3,
167        "idea_info": 31.84697995765805,
168        "idea_late_eta": 2.716472166477494,
169        "idea_rollbacks": 0
170      },
171      {
172        "seed": 4,
173        "idea_info": 43.19129275120528,
174        "idea_late_eta": 0.4035124075587761,
175        "idea_rollbacks": 0
176      },
177      {
178        "seed": 5,
179        "idea_info": 33.22536463027882,
180        "idea_late_eta": -2.2281370987865134,
181        "idea_rollbacks": 2
182      },
183      {
184        "seed": 6,
185        "idea_info": 33.59606871616029,
186        "idea_late_eta": 0.8177034771039023,
187        "idea_rollbacks": 0
188      },
189      {
190        "seed": 7,
191        "idea_info": 36.66021512586849,
192        "idea_late_eta": 0.668399072113469,
193        "idea_rollbacks": 2
194      }
195    ]
196  },
197  "protocol_notes": {
198    "n_train": 2000,
199    "n_test": 500,
200    "epochs": 15,
201    "common_validation_fraction": 0.2,
202    "track_match": "tabular is the built-in optimizer/scheduler track"
203  }
204}