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    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.0015
 15        },
 16        "mean": 0.0019174260669387877
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.0009721333481138572
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.006
 27        },
 28        "mean": 0.0007438920001732185
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.0007792052056174725,
 33      "std": 0.00021144133531208117,
 34      "per_seed": [
 35        0.0007514121243730187,
 36        0.0005835454212501645,
 37        0.0005766521790064871,
 38        0.0010639582760632038,
 39        0.0008014270570129156,
 40        0.0006233751191757619,
 41        0.0006578490138053894,
 42        0.001175422454252839
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.024256352335214615,
 49    "std": 0.0038067872943047523,
 50    "per_seed": [
 51      0.021198168396949768,
 52      0.02363172546029091,
 53      0.030124139040708542,
 54      0.02928881347179413,
 55      0.02132338099181652,
 56      0.018241815268993378,
 57      0.024838322773575783,
 58      0.02540445327758789
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.023477147129597142,
 64    "idea_wins": 0,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      0.02044675627257675,
 68      0.023048180039040744,
 69      0.029547486861702055,
 70      0.028224855195730925,
 71      0.020521953934803605,
 72      0.017618440149817616,
 73      0.024180473759770393,
 74      0.02422903082333505
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 0.0033576493476355656,
 78    "mde_rel_pct": 430.9069451063066,
 79    "verdict": "idea worse (significant)",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "idea_config": {
 84      "lr": 0.003
 85    },
 86    "idea_sweep": [
 87      {
 88        "cfg": {
 89          "lr": 0.0015
 90        },
 91        "result": {
 92          "mean": 0.1092000212520361,
 93          "std": 0.032826994153617124,
 94          "per_seed": [
 95            0.12505441904067993,
 96            0.13431435823440552,
 97            0.10349666327238083,
 98            0.1606961041688919,
 99            0.060887403786182404,
100            0.12317634373903275,
101            0.05900014936923981,
102            0.10697472840547562
103          ],
104          "n": 8
105        }
106      },
107      {
108        "cfg": {
109          "lr": 0.003
110        },
111        "result": {
112          "mean": 0.024256352335214615,
113          "std": 0.0038067872943047523,
114          "per_seed": [
115            0.021198168396949768,
116            0.02363172546029091,
117            0.030124139040708542,
118            0.02928881347179413,
119            0.02132338099181652,
120            0.018241815268993378,
121            0.024838322773575783,
122            0.02540445327758789
123          ],
124          "n": 8
125        }
126      },
127      {
128        "cfg": {
129          "lr": 0.006
130        },
131        "result": {
132          "mean": 0.028757888358086348,
133          "std": 0.005178806657688538,
134          "per_seed": [
135            0.021722447127103806,
136            0.029085084795951843,
137            0.03460188955068588,
138            0.038650788366794586,
139            0.024053702130913734,
140            0.02563602291047573,
141            0.02774704061448574,
142            0.028566131368279457
143          ],
144          "n": 8
145        }
146      }
147    ],
148    "mechanism_signature": {
149      "prediction": "direct topology interpolation can transiently violate calibration accuracy; exact receding-horizon filtering accepts no violating step",
150      "trained_baseline_unfiltered_violations_of_9": 7,
151      "trained_idea_accepted_step_violations": 0,
152      "trained_idea_rejected_candidates": 55,
153      "trained_idea_transition_steps": 1,
154      "predicted_zero_accepted_violations": true,
155      "confirmed": true
156    },
157    "epsilon_calibration_mse": 0.02,
158    "custom_track": null
159  },
160  "protocol": {
161    "paired_seeds": [
162      0,
163      1,
164      2,
165      3,
166      4,
167      5,
168      6,
169      7
170    ],
171    "baseline_grid": [
172      {
173        "lr": 0.0015
174      },
175      {
176        "lr": 0.003
177      },
178      {
179        "lr": 0.006
180      }
181    ],
182    "idea_grid": [
183      {
184        "lr": 0.0015
185      },
186      {
187        "lr": 0.003
188      },
189      {
190        "lr": 0.006
191      }
192    ],
193    "epochs": 18,
194    "n_train": 400,
195    "n_test": 200
196  }
197}