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": 0.0008724518556846306
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.00035974886850453913
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 4.963686433256953e-05
 29      }
 30    ],
 31    "full": {
 32      "mean": 4.311938073442434e-05,
 33      "std": 2.1245040862825782e-05,
 34      "per_seed": [
 35        9.722933464217931e-05,
 36        3.783951615332626e-05,
 37        3.003067286044825e-05,
 38        3.3447933674324304e-05,
 39        4.86072531202808e-05,
 40        2.836398380168248e-05,
 41        3.394361920072697e-05,
 42        3.549273242242634e-05
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 4.2328419567638775e-05,
 49    "std": 2.1386848202922675e-05,
 50    "per_seed": [
 51      9.703257819637656e-05,
 52      3.75409254047554e-05,
 53      2.9079366868245415e-05,
 54      3.262508107582107e-05,
 55      4.653289943234995e-05,
 56      2.7314470571582206e-05,
 57      3.378422115929425e-05,
 58      3.471781383268535e-05
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": -7.909611667855643e-07,
 64    "idea_wins": 8,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      -1.967564458027482e-07,
 68      -2.9859074857085943e-07,
 69      -9.513059922028333e-07,
 70      -8.22852598503232e-07,
 71      -2.074353687930852e-06,
 72      -1.049513230100274e-06,
 73      -1.593980414327234e-07,
 74      -7.749185897409916e-07
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 5.226854739698207e-07,
 78    "mde_rel_pct": 1.2121822370063282,
 79    "verdict": "idea better (significant)",
 80    "system_worked": true
 81  },
 82  "mechanism_signature": {
 83    "prediction": "orthogonal transitions preserve hidden norms and patch loss reduces disagreement",
 84    "predicted_norm_error": 0.0,
 85    "observed_max_norm_error": 2.384185791015625e-07,
 86    "predicted_disagreement_change": "negative",
 87    "observed_baseline_disagreement": 0.002737239032285288,
 88    "observed_idea_disagreement": 0.002110361514496617,
 89    "confirmed": true,
 90    "math_sanity": {
 91      "max_norm_error": 4.76837158203125e-07,
 92      "predicted": "orthogonal gauge preserves hidden norm"
 93    },
 94    "trained_model_behavior": {
 95      "baseline": [
 96        {
 97          "metric": 9.722933464217931e-05,
 98          "disagreement": 0.0028517728205770254,
 99          "norm_error": 0.0,
100          "angle_mean_abs": 0.0
101        },
102        {
103          "metric": 3.783951615332626e-05,
104          "disagreement": 0.0026706315111368895,
105          "norm_error": 0.0,
106          "angle_mean_abs": 0.0
107        },
108        {
109          "metric": 3.003067286044825e-05,
110          "disagreement": 0.0027825413271784782,
111          "norm_error": 0.0,
112          "angle_mean_abs": 0.0
113        },
114        {
115          "metric": 3.3447933674324304e-05,
116          "disagreement": 0.002656190888956189,
117          "norm_error": 0.0,
118          "angle_mean_abs": 0.0
119        },
120        {
121          "metric": 4.86072531202808e-05,
122          "disagreement": 0.002491437830030918,
123          "norm_error": 0.0,
124          "angle_mean_abs": 0.0
125        },
126        {
127          "metric": 2.836398380168248e-05,
128          "disagreement": 0.003299158299341798,
129          "norm_error": 0.0,
130          "angle_mean_abs": 0.0
131        },
132        {
133          "metric": 3.394361920072697e-05,
134          "disagreement": 0.0026531326584517956,
135          "norm_error": 0.0,
136          "angle_mean_abs": 0.0
137        },
138        {
139          "metric": 3.549273242242634e-05,
140          "disagreement": 0.00249304692260921,
141          "norm_error": 0.0,
142          "angle_mean_abs": 0.0
143        }
144      ],
145      "idea": [
146        {
147          "metric": 9.703257819637656e-05,
148          "disagreement": 0.0022101164795458317,
149          "norm_error": 2.384185791015625e-07,
150          "angle_mean_abs": 0.05565014109015465
151        },
152        {
153          "metric": 3.75409254047554e-05,
154          "disagreement": 0.0020225902553647757,
155          "norm_error": 2.384185791015625e-07,
156          "angle_mean_abs": 0.06866727769374847
157        },
158        {
159          "metric": 2.9079366868245415e-05,
160          "disagreement": 0.002208986319601536,
161          "norm_error": 2.384185791015625e-07,
162          "angle_mean_abs": 0.05578428506851196
163        },
164        {
165          "metric": 3.262508107582107e-05,
166          "disagreement": 0.0020868959836661816,
167          "norm_error": 2.384185791015625e-07,
168          "angle_mean_abs": 0.048349007964134216
169        },
170        {
171          "metric": 4.653289943234995e-05,
172          "disagreement": 0.001802699756808579,
173          "norm_error": 2.384185791015625e-07,
174          "angle_mean_abs": 0.053223490715026855
175        },
176        {
177          "metric": 2.7314470571582206e-05,
178          "disagreement": 0.0024720188230276108,
179          "norm_error": 2.384185791015625e-07,
180          "angle_mean_abs": 0.056646957993507385
181        },
182        {
183          "metric": 3.378422115929425e-05,
184          "disagreement": 0.002076399978250265,
185          "norm_error": 2.384185791015625e-07,
186          "angle_mean_abs": 0.05470898002386093
187        },
188        {
189          "metric": 3.471781383268535e-05,
190          "disagreement": 0.0020031845197081566,
191          "norm_error": 2.384185791015625e-07,
192          "angle_mean_abs": 0.054334621876478195
193        }
194      ]
195    },
196    "idea_sweep": [
197      {
198        "cfg": {
199          "lr": 0.001,
200          "lambda_patch": 0.01
201        },
202        "result": {
203          "mean": 0.0009701697272248566,
204          "std": 0.0002551454195727328,
205          "per_seed": [
206            0.0006990438560023904,
207            0.0009627738618291914,
208            0.0007161990506574512,
209            0.0011117245303466916,
210            0.0007498612976633012,
211            0.0008323467336595058,
212            0.001259242300875485,
213            0.0014301661867648363
214          ],
215          "n": 8
216        }
217      },
218      {
219        "cfg": {
220          "lr": 0.003,
221          "lambda_patch": 0.01
222        },
223        "result": {
224          "mean": 0.0003214057542209048,
225          "std": 5.179809947448413e-05,
226          "per_seed": [
227            0.0003400232526473701,
228            0.0004083190578967333,
229            0.00031012375256977975,
230            0.00038058325299061835,
231            0.0003361486305948347,
232            0.00027818765374831855,
233            0.0002542929141782224,
234            0.0002635675191413611
235          ],
236          "n": 8
237        }
238      },
239      {
240        "cfg": {
241          "lr": 0.01,
242          "lambda_patch": 0.01
243        },
244        "result": {
245          "mean": 4.2328419567638775e-05,
246          "std": 2.1386848202922675e-05,
247          "per_seed": [
248            9.703257819637656e-05,
249            3.75409254047554e-05,
250            2.9079366868245415e-05,
251            3.262508107582107e-05,
252            4.653289943234995e-05,
253            2.7314470571582206e-05,
254            3.378422115929425e-05,
255            3.471781383268535e-05
256          ],
257          "n": 8
258        }
259      }
260    ],
261    "track_rationale": "Dynamics is the built-in stability/control track; local temporal patches are neighboring state windows."
262  }
263}