Carrier-Probed Hidden-State Training / bench_report.json

Failed on benchmark

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": 0.01
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.005440622277092189
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.0012262430100236088
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.000702349963830784
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.0006528829871967901,
 33      "std": 0.0001692890895210604,
 34      "per_seed": [
 35        0.0006883637397550046,
 36        0.00054119125707075,
 37        0.0005475711077451706,
 38        0.0010322737507522106,
 39        0.00068665825529024,
 40        0.00041018708725459874,
 41        0.0006441730074584484,
 42        0.0006726456922478974
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.000652743416139856,
 49    "std": 0.0001692304843358837,
 50    "per_seed": [
 51      0.0006887014023959637,
 52      0.0005411907332018018,
 53      0.0005468056187964976,
 54      0.0010318164713680744,
 55      0.0006858076085336506,
 56      0.00041013583540916443,
 57      0.0006440518773160875,
 58      0.0006734377820976079
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": -1.3957105693407357e-07,
 64    "idea_wins": 6,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      3.3766264095902443e-07,
 68      -5.238689482212067e-10,
 69      -7.654889486730099e-07,
 70      -4.5727938413619995e-07,
 71      -8.506467565894127e-07,
 72      -5.125184543430805e-08,
 73      -1.2113014236092567e-07,
 74      7.920898497104645e-07
 75    ],
 76    "p_value": 0.45895,
 77    "mde": 4.6032512964596226e-07,
 78    "mde_rel_pct": 0.07050652853161457,
 79    "verdict": "no measurable effect",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "mechanism_signature": {
 84      "carrier_amp": 0.3,
 85      "predicted_abs_delta_mean": 0.0008173562528099865,
 86      "observed_nominal_abs_delta_mean": 5.861693352926522e-10,
 87      "predicted_observed_correlation_mean": NaN,
 88      "quadratic_prediction": "not confirmed at NN scale: only one amplitude was behaviorally tested",
 89      "confirmed": false
 90    },
 91    "idea_sweep": [
 92      {
 93        "amp": 0.05,
 94        "result": {
 95          "mean": 0.0006528785997943487,
 96          "std": 0.0001692863867624003,
 97          "per_seed": [
 98            0.0006883704336360097,
 99            0.0005411913734860718,
100            0.0005475485231727362,
101            0.001032259315252304,
102            0.000686634739395231,
103            0.00041018894989974797,
104            0.0006441685254685581,
105            0.0006726669380441308
106          ],
107          "n": 8
108        }
109      },
110      {
111        "amp": 0.15,
112        "result": {
113          "mean": 0.0006528485828312114,
114          "std": 0.00016926723045994872,
115          "per_seed": [
116            0.000688436848577112,
117            0.0005411922466009855,
118            0.0005473769851960242,
119            0.0010321518639102578,
120            0.0006864455062896013,
121            0.0004102009115740657,
122            0.0006441408768296242,
123            0.0006728434236720204
124          ],
125          "n": 8
126        }
127      },
128      {
129        "amp": 0.3,
130        "result": {
131          "mean": 0.000652743416139856,
132          "std": 0.0001692304843358837,
133          "per_seed": [
134            0.0006887014023959637,
135            0.0005411907332018018,
136            0.0005468056187964976,
137            0.0010318164713680744,
138            0.0006858076085336506,
139            0.00041013583540916443,
140            0.0006440518773160875,
141            0.0006734377820976079
142          ],
143          "n": 8
144        }
145      }
146    ],
147    "track_justification": "Dynamics matches the proposed hidden-state observability/reachability and stability/control structure."
148  },
149  "idea_sweep": [
150    {
151      "amp": 0.05,
152      "result": {
153        "mean": 0.0006528785997943487,
154        "std": 0.0001692863867624003,
155        "per_seed": [
156          0.0006883704336360097,
157          0.0005411913734860718,
158          0.0005475485231727362,
159          0.001032259315252304,
160          0.000686634739395231,
161          0.00041018894989974797,
162          0.0006441685254685581,
163          0.0006726669380441308
164        ],
165        "n": 8
166      }
167    },
168    {
169      "amp": 0.15,
170      "result": {
171        "mean": 0.0006528485828312114,
172        "std": 0.00016926723045994872,
173        "per_seed": [
174          0.000688436848577112,
175          0.0005411922466009855,
176          0.0005473769851960242,
177          0.0010321518639102578,
178          0.0006864455062896013,
179          0.0004102009115740657,
180          0.0006441408768296242,
181          0.0006728434236720204
182        ],
183        "n": 8
184      }
185    },
186    {
187      "amp": 0.3,
188      "result": {
189        "mean": 0.000652743416139856,
190        "std": 0.0001692304843358837,
191        "per_seed": [
192          0.0006887014023959637,
193          0.0005411907332018018,
194          0.0005468056187964976,
195          0.0010318164713680744,
196          0.0006858076085336506,
197          0.00041013583540916443,
198          0.0006440518773160875,
199          0.0006734377820976079
200        ],
201        "n": 8
202      }
203    }
204  ]
205}