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": 0.001
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 140.784441947937
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 870.165714263916
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 682.9689865112305
 29      }
 30    ],
 31    "full": {
 32      "mean": 118.6587815284729,
 33      "std": 53.34609536660658,
 34      "per_seed": [
 35        202.70230102539062,
 36        60.04500961303711,
 37        118.25131225585938,
 38        182.13914489746094,
 39        153.9148406982422,
 40        100.01561737060547,
 41        89.7911148071289,
 42        42.410911560058594
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.5461385995149612,
 49    "std": 0.43340968114270667,
 50    "per_seed": [
 51      0.30781564116477966,
 52      0.6033635139465332,
 53      0.3968035876750946,
 54      1.6646853685379028,
 55      0.3090420663356781,
 56      0.2753473222255707,
 57      0.3914429545402527,
 58      0.4206083416938782
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": -118.11264292895794,
 64    "idea_wins": 8,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      -202.39448538422585,
 68      -59.441646099090576,
 69      -117.85450866818428,
 70      -180.47445952892303,
 71      -153.6057986319065,
 72      -99.7402700483799,
 73      -89.39967185258865,
 74      -41.990303218364716
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 47.5644061392987,
 78    "mde_rel_pct": 40.085028285820826,
 79    "verdict": "idea better (significant)",
 80    "system_worked": true
 81  },
 82  "audit": {
 83    "structural_match": "dynamics stability/control",
 84    "base_architecture": "rnn_small GRU plus scalar Laurent readout",
 85    "idea_configs": [
 86      {
 87        "lr": 0.001
 88      },
 89      {
 90        "lr": 0.003
 91      },
 92      {
 93        "lr": 0.01
 94      }
 95    ],
 96    "baseline_grid": [
 97      {
 98        "lr": 0.001
 99      },
100      {
101        "lr": 0.003
102      },
103      {
104        "lr": 0.01
105      }
106    ],
107    "baseline_best_lr": 0.001,
108    "idea_best_lr": 0.001,
109    "epochs": 12,
110    "n_train": 400,
111    "n_test": 200,
112    "parameter_parity": true
113  },
114  "mechanism_signature": {
115    "prediction": "unsafe output scales as t^-2; safe output is bounded (t^0)",
116    "observed": {
117      "baseline": {
118        "distances": [
119          0.05,
120          0.03,
121          0.02,
122          0.01
123        ],
124        "mean_abs_outputs": [
125          1.7172616720199585,
126          4.7458672523498535,
127          10.651176452636719,
128          42.497459411621094
129        ],
130        "loglog_slope": -1.99381250167288,
131        "predicted_slope": -2.0,
132        "absolute_slope_error": 0.006187498327119956
133      },
134      "idea": {
135        "distances": [
136          0.05,
137          0.03,
138          0.02,
139          0.01
140        ],
141        "mean_abs_outputs": [
142          1.0119659900665283,
143          1.0068100690841675,
144          1.0042619705200195,
145          1.0017340183258057
146        ],
147        "loglog_slope": 0.00618747125927932,
148        "predicted_slope": 0.0,
149        "absolute_slope_error": 0.00618747125927932
150      }
151    },
152    "confirmed": true
153  },
154  "idea_sweep": [
155    {
156      "cfg": {
157        "lr": 0.001
158      },
159      "mean": 0.5461385995149612
160    },
161    {
162      "cfg": {
163        "lr": 0.003
164      },
165      "mean": 0.6001838333904743
166    },
167    {
168      "cfg": {
169        "lr": 0.01
170      },
171      "mean": 0.9318437203764915
172    }
173  ]
174}