Drift-Balanced Adaptive Constraint Multiplier / bench_report.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "dynamics",
  4  "model": "rnn_small",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.006,
 10      "penalty": 0.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "penalty": 0.0
 17        },
 18        "mean": 0.0012014944513794035
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "penalty": 0.03
 24        },
 25        "mean": 0.0013431779225356877
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.001,
 30          "penalty": 0.1
 31        },
 32        "mean": 0.001977995882043615
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.001,
 37          "penalty": 0.3
 38        },
 39        "mean": 0.005485758418217301
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.001,
 44          "penalty": 1.0
 45        },
 46        "mean": 0.025594600941985846
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.003,
 51          "penalty": 0.0
 52        },
 53        "mean": 0.0006355914229061455
 54      },
 55      {
 56        "cfg": {
 57          "lr": 0.003,
 58          "penalty": 0.03
 59        },
 60        "mean": 0.0007585847488371655
 61      },
 62      {
 63        "cfg": {
 64          "lr": 0.003,
 65          "penalty": 0.1
 66        },
 67        "mean": 0.0014228892396204174
 68      },
 69      {
 70        "cfg": {
 71          "lr": 0.003,
 72          "penalty": 0.3
 73        },
 74        "mean": 0.00530714844353497
 75      },
 76      {
 77        "cfg": {
 78          "lr": 0.003,
 79          "penalty": 1.0
 80        },
 81        "mean": 0.031064636074006557
 82      },
 83      {
 84        "cfg": {
 85          "lr": 0.006,
 86          "penalty": 0.0
 87        },
 88        "mean": 0.0003905312332790345
 89      },
 90      {
 91        "cfg": {
 92          "lr": 0.006,
 93          "penalty": 0.03
 94        },
 95        "mean": 0.0005399620640673675
 96      },
 97      {
 98        "cfg": {
 99          "lr": 0.006,
100          "penalty": 0.1
101        },
102        "mean": 0.0013079957861918956
103      },
104      {
105        "cfg": {
106          "lr": 0.006,
107          "penalty": 0.3
108        },
109        "mean": 0.005471856915391982
110      },
111      {
112        "cfg": {
113          "lr": 0.006,
114          "penalty": 1.0
115        },
116        "mean": 0.03726070327684283
117      }
118    ],
119    "full": {
120      "mean": 0.00038071037488407455,
121      "std": 7.927794626057588e-05,
122      "per_seed": [
123        0.0003860906872432679,
124        0.0004608586023095995,
125        0.00025328659103251994,
126        0.00046188905253075063,
127        0.00042785078403539956,
128        0.00024962733732536435,
129        0.00042033527279272676,
130        0.0003857446718029678
131      ],
132      "n": 8
133    }
134  },
135  "idea": {
136    "mean": 0.0021949997608317062,
137    "std": 0.0004374952005319115,
138    "per_seed": [
139      0.0026410226710140705,
140      0.0018085994524881244,
141      0.00271115405485034,
142      0.0028186936397105455,
143      0.001880061230622232,
144      0.0017161071300506592,
145      0.0017505508149042726,
146      0.002233809093013406
147    ],
148    "n": 8
149  },
150  "comparison": {
151    "delta_mean": 0.0018142893859476317,
152    "idea_wins": 0,
153    "n_pairs": 8,
154    "per_seed_diffs": [
155      0.0022549319837708026,
156      0.0013477408501785249,
157      0.00245786746381782,
158      0.002356804587179795,
159      0.0014522104465868324,
160      0.0014664797927252948,
161      0.0013302155421115458,
162      0.001848064421210438
163    ],
164    "p_value": 0.0081,
165    "mde": 0.00040074238262426,
166    "mde_rel_pct": 105.26174463889649,
167    "verdict": "idea worse (significant)",
168    "system_worked": false
169  },
170  "selected_idea_cfg": {
171    "lr": 0.006,
172    "alpha": 0.02
173  },
174  "math_check": {
175    "max_projection_error": 0.0,
176    "interior_drift_error": 5.551115123125783e-17,
177    "constant_positive_drift_cap_steps": 63
178  },
179  "track_justification": "Dynamics is structurally matched: controlled pendulum rollout and terminal feasibility.",
180  "mechanism_signature": {
181    "formula": "delta_lambda = alpha*(vbar-tau) away from projection",
182    "per_seed": [
183      {
184        "seed": 0,
185        "mean_violation": 0.0825553372502327,
186        "mean_lambda": 0.20099468181282282,
187        "drift_prediction_mae": 4.7076746711652336e-18,
188        "drift_observed": 0.001723537926399518,
189        "drift_predicted": 0.0017235379263995185,
190        "cap_fraction": 0.0
191      },
192      {
193        "seed": 1,
194        "mean_violation": 0.0704090990126133,
195        "mean_lambda": 0.17491563167423013,
196        "drift_prediction_mae": 3.7114563654548425e-18,
197        "drift_observed": 0.001494265170620074,
198        "drift_predicted": 0.001494265170620074,
199        "cap_fraction": 0.0
200      },
201      {
202        "seed": 2,
203        "mean_violation": 0.07129735685884953,
204        "mean_lambda": 0.16706302325800054,
205        "drift_prediction_mae": 4.001007183741646e-18,
206        "drift_observed": 0.0014390677878899228,
207        "drift_predicted": 0.0014390677878899233,
208        "cap_fraction": 0.0
209      },
210      {
211        "seed": 3,
212        "mean_violation": 0.06800611577928066,
213        "mean_lambda": 0.16455769450590002,
214        "drift_prediction_mae": 3.9042034183411545e-18,
215        "drift_observed": 0.0014052169518693086,
216        "drift_predicted": 0.0014052169518693096,
217        "cap_fraction": 0.0
218      },
219      {
220        "seed": 4,
221        "mean_violation": 0.077877695299685,
222        "mean_lambda": 0.1931413715332746,
223        "drift_prediction_mae": 4.34713442491807e-18,
224        "drift_observed": 0.0016449038421232547,
225        "drift_predicted": 0.0016449038421232547,
226        "cap_fraction": 0.0
227      },
228      {
229        "seed": 5,
230        "mean_violation": 0.0745898948982358,
231        "mean_lambda": 0.19240893951803442,
232        "drift_prediction_mae": 4.721442317799971e-18,
233        "drift_observed": 0.0016378540495440126,
234        "drift_predicted": 0.001637854049544013,
235        "cap_fraction": 0.0
236      },
237      {
238        "seed": 6,
239        "mean_violation": 0.07793739642947912,
240        "mean_lambda": 0.1956254186630249,
241        "drift_prediction_mae": 4.275714758000374e-18,
242        "drift_observed": 0.0016655272277929485,
243        "drift_predicted": 0.0016655272277929481,
244        "cap_fraction": 0.0
245      },
246      {
247        "seed": 7,
248        "mean_violation": 0.06042268052697182,
249        "mean_lambda": 0.15612414248287676,
250        "drift_prediction_mae": 3.4690167129962788e-18,
251        "drift_observed": 0.0013308829530363985,
252        "drift_predicted": 0.001330882953036399,
253        "cap_fraction": 0.0
254      }
255    ],
256    "observed_mean_drift": 0.0015426569886594298,
257    "predicted_mean_drift": 0.00154265698865943,
258    "mean_drift_prediction_mae": 4.142206231552196e-18,
259    "confirmed": true
260  }
261}