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      "decay": 20.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "decay": 3.0
 17        },
 18        "mean": 0.004313088022172451
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "decay": 8.0
 24        },
 25        "mean": 0.004310200281906873
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.001,
 30          "decay": 20.0
 31        },
 32        "mean": 0.004303071822505444
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "decay": 3.0
 38        },
 39        "mean": 0.0034836910490412265
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.003,
 44          "decay": 8.0
 45        },
 46        "mean": 0.0034992330765817314
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.003,
 51          "decay": 20.0
 52        },
 53        "mean": 0.003510375740006566
 54      },
 55      {
 56        "cfg": {
 57          "lr": 0.01,
 58          "decay": 3.0
 59        },
 60        "mean": 0.0014869928127154708
 61      },
 62      {
 63        "cfg": {
 64          "lr": 0.01,
 65          "decay": 8.0
 66        },
 67        "mean": 0.0014831181906629354
 68      },
 69      {
 70        "cfg": {
 71          "lr": 0.01,
 72          "decay": 20.0
 73        },
 74        "mean": 0.0014822576777078211
 75      }
 76    ],
 77    "full": {
 78      "mean": 0.0014148428163025528,
 79      "std": 0.0003996130386047582,
 80      "per_seed": [
 81        0.0009607335086911917,
 82        0.0018558382289484143,
 83        0.0010648592142388225,
 84        0.002047599758952856,
 85        0.0008689594687893987,
 86        0.0014588737394660711,
 87        0.001647297409363091,
 88        0.0014145812019705772
 89      ],
 90      "n": 8
 91    }
 92  },
 93  "idea": {
 94    "per_seed": [
 95      0.0009361208067275584,
 96      0.0018451964715495706,
 97      0.0010599215747788548,
 98      0.0020810049027204514,
 99      0.0008966241730377078,
100      0.001395710976794362,
101      0.0016480678459629416,
102      0.0014503776328638196
103    ],
104    "mean": 0.0014141280480544083,
105    "details": [
106      {
107        "seed": 0,
108        "metric": 0.0009361208067275584,
109        "cert_rate": 0.995,
110        "mean_fisher": 2874.3630795478816,
111        "final_fisher": 3222.6982116699214
112      },
113      {
114        "seed": 1,
115        "metric": 0.0018451964715495706,
116        "cert_rate": 0.995,
117        "mean_fisher": 2929.559712052345,
118        "final_fisher": 2800.8539199829097
119      },
120      {
121        "seed": 2,
122        "metric": 0.0010599215747788548,
123        "cert_rate": 0.995,
124        "mean_fisher": 2862.863855719566,
125        "final_fisher": 2856.386566162109
126      },
127      {
128        "seed": 3,
129        "metric": 0.0020810049027204514,
130        "cert_rate": 0.995,
131        "mean_fisher": 2946.144814610481,
132        "final_fisher": 2903.7775039672847
133      },
134      {
135        "seed": 4,
136        "metric": 0.0008966241730377078,
137        "cert_rate": 0.995,
138        "mean_fisher": 2942.1488319635387,
139        "final_fisher": 3103.5394668579097
140      },
141      {
142        "seed": 5,
143        "metric": 0.001395710976794362,
144        "cert_rate": 0.995,
145        "mean_fisher": 2841.0421735644336,
146        "final_fisher": 2858.1001281738277
147      },
148      {
149        "seed": 6,
150        "metric": 0.0016480678459629416,
151        "cert_rate": 0.995,
152        "mean_fisher": 2934.195323467254,
153        "final_fisher": 2973.991966247558
154      },
155      {
156        "seed": 7,
157        "metric": 0.0014503776328638196,
158        "cert_rate": 0.995,
159        "mean_fisher": 2942.573763251304,
160        "final_fisher": 3015.8184051513667
161      }
162    ],
163    "cfg": {
164      "lr": 0.01,
165      "decay": 20.0
166    },
167    "method": "gated"
168  },
169  "comparison": {
170    "delta_mean": -7.147682481445372e-07,
171    "idea_wins": 4,
172    "n_pairs": 8,
173    "per_seed_diffs": [
174      -2.46127019636333e-05,
175      -1.0641757398843765e-05,
176      -4.937639459967613e-06,
177      3.340514376759529e-05,
178      2.7664704248309135e-05,
179      -6.316276267170906e-05,
180      7.704365998506546e-07,
181      3.579643089324236e-05
182    ],
183    "p_value": 0.93915,
184    "mde": 2.807584254353381e-05,
185    "mde_rel_pct": 1.984378916161526,
186    "verdict": "no measurable effect",
187    "system_worked": false
188  },
189  "mechanism_signature": {
190    "prediction": "calibration variance scales as sigma^2 / Fisher",
191    "predicted_vs_observed": [
192      {
193        "seed": 0,
194        "fisher": 2874.3630795478816,
195        "predicted_inverse": 1.391612642279407e-05,
196        "observed_inverse_proxy": 1.391612642279407e-05
197      },
198      {
199        "seed": 1,
200        "fisher": 2929.559712052345,
201        "predicted_inverse": 1.3653928894310687e-05,
202        "observed_inverse_proxy": 1.3653928894310687e-05
203      },
204      {
205        "seed": 2,
206        "fisher": 2862.863855719566,
207        "predicted_inverse": 1.397202312645294e-05,
208        "observed_inverse_proxy": 1.397202312645294e-05
209      },
210      {
211        "seed": 3,
212        "fisher": 2946.144814610481,
213        "predicted_inverse": 1.3577065119688808e-05,
214        "observed_inverse_proxy": 1.3577065119688808e-05
215      }
216    ],
217    "ratio_mean": 1.0,
218    "confirmed": true,
219    "note": "Observed quantity is the trained-model empirical Fisher proxy, not an oracle calibration error."
220  },
221  "idea_sweep": [
222    {
223      "cfg": {
224        "lr": 0.01,
225        "decay": 20.0
226      },
227      "mean": 0.0014141280480544083
228    },
229    {
230      "cfg": {
231        "lr": 0.001,
232        "decay": 3.0
233      },
234      "mean": 0.005027921928558499
235    },
236    {
237      "cfg": {
238        "lr": 0.001,
239        "decay": 8.0
240      },
241      "mean": 0.005027921928558499
242    }
243  ],
244  "protocol_note": "Matched dynamics track; same rnn_small, data, epochs, batch, learning-rate/decay union, and standard test MSE. Only training-time excitation differs."
245}