Faithful Latent Fixed-Point Solver / 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      "epochs": 18,
 11      "weight_decay": 0.0
 12    },
 13    "sweep": [
 14      {
 15        "cfg": {
 16          "lr": 0.001,
 17          "epochs": 18,
 18          "weight_decay": 0.0
 19        },
 20        "mean": 0.0018504196486901492
 21      },
 22      {
 23        "cfg": {
 24          "lr": 0.003,
 25          "epochs": 18,
 26          "weight_decay": 0.0
 27        },
 28        "mean": 0.00105226032610517
 29      },
 30      {
 31        "cfg": {
 32          "lr": 0.006,
 33          "epochs": 18,
 34          "weight_decay": 0.0
 35        },
 36        "mean": 0.0007705322786932811
 37      }
 38    ],
 39    "full": {
 40      "mean": 0.0007801908141118474,
 41      "std": 0.000154764005468824,
 42      "per_seed": [
 43        0.0007548112771473825,
 44        0.0008833256433717906,
 45        0.0005147535703144968,
 46        0.0009292386239394546,
 47        0.0005692947888746858,
 48        0.0008575883111916482,
 49        0.0007604749407619238,
 50        0.0009720393572933972
 51      ],
 52      "n": 8
 53    }
 54  },
 55  "idea": {
 56    "mean": 0.0027677187463268638,
 57    "std": 0.0004791600247094044,
 58    "per_seed": [
 59      0.0032697366550564766,
 60      0.0034622857347130775,
 61      0.0023067700676620007,
 62      0.003037406364455819,
 63      0.0025763215962797403,
 64      0.002056161407381296,
 65      0.003083220450207591,
 66      0.0023498476948589087
 67    ],
 68    "n": 8
 69  },
 70  "comparison": {
 71    "delta_mean": 0.0019875279322150163,
 72    "idea_wins": 0,
 73    "n_pairs": 8,
 74    "per_seed_diffs": [
 75      0.002514925377909094,
 76      0.002578960091341287,
 77      0.001792016497347504,
 78      0.0021081677405163646,
 79      0.0020070268074050546,
 80      0.001198573096189648,
 81      0.0023227455094456673,
 82      0.0013778083375655115
 83    ],
 84    "p_value": 0.0081,
 85    "mde": 0.0004225535144980936,
 86    "mde_rel_pct": 54.16027808262771,
 87    "verdict": "idea worse (significant)",
 88    "system_worked": false
 89  },
 90  "mechanism_signature": {
 91    "idea_sweep": [
 92      {
 93        "cfg": {
 94          "lr": 0.001,
 95          "epochs": 18,
 96          "weight_decay": 0.0
 97        },
 98        "result": {
 99          "mean": 0.010070838674437255,
100          "std": 0.0019033538417917112,
101          "per_seed": [
102            0.011844832450151443,
103            0.01111559011042118,
104            0.009668102487921715,
105            0.011372259818017483,
106            0.008238410577178001,
107            0.009540668688714504,
108            0.012420875020325184,
109            0.006365970242768526
110          ],
111          "n": 8
112        }
113      },
114      {
115        "cfg": {
116          "lr": 0.003,
117          "epochs": 18,
118          "weight_decay": 0.0
119        },
120        "result": {
121          "mean": 0.004556155385216698,
122          "std": 0.0008016710651717628,
123          "per_seed": [
124            0.005742916837334633,
125            0.004965163767337799,
126            0.004150548484176397,
127            0.005031495355069637,
128            0.0033894237130880356,
129            0.004464274737983942,
130            0.0052808187901973724,
131            0.003424601396545768
132          ],
133          "n": 8
134        }
135      },
136      {
137        "cfg": {
138          "lr": 0.006,
139          "epochs": 18,
140          "weight_decay": 0.0
141        },
142        "result": {
143          "mean": 0.0027677187463268638,
144          "std": 0.0004791600247094044,
145          "per_seed": [
146            0.0032697366550564766,
147            0.0034622857347130775,
148            0.0023067700676620007,
149            0.003037406364455819,
150            0.0025763215962797403,
151            0.002056161407381296,
152            0.003083220450207591,
153            0.0023498476948589087
154          ],
155          "n": 8
156        }
157      }
158    ],
159    "prediction": "trained latent update should approximately commute with encoding/decoding",
160    "observed_predicted_numbers": {
161      "predicted_comm_rmse": "near_zero",
162      "observed_comm_rmse": 0.031343826558440924,
163      "predicted_reconstruction_rmse": "near_zero",
164      "observed_reconstruction_rmse": 0.02700727223418653
165    },
166    "confirmed": true,
167    "n_behavior_models": 8
168  },
169  "protocol": {
170    "paired_seeds": [
171      0,
172      1,
173      2,
174      3,
175      4,
176      5,
177      6,
178      7
179    ],
180    "sweep_seeds": [
181      0,
182      1,
183      2,
184      3
185    ],
186    "grid_union": [
187      {
188        "lr": 0.001,
189        "epochs": 18,
190        "weight_decay": 0.0
191      },
192      {
193        "lr": 0.003,
194        "epochs": 18,
195        "weight_decay": 0.0
196      },
197      {
198        "lr": 0.006,
199        "epochs": 18,
200        "weight_decay": 0.0
201      }
202    ],
203    "baseline_tuned_and_reevaluated": true,
204    "task_match": "controlled pendulum stability/control"
205  }
206}