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": 0.001864717691205442
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.0010901302302954718
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.0006186682076076977
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.00054517226817552,
 33      "std": 0.00020361842727135583,
 34      "per_seed": [
 35        0.0004930912400595844,
 36        0.0004765945195686072,
 37        0.0004673648509196937,
 38        0.0010376222198829055,
 39        0.000511663849465549,
 40        0.00027233464061282575,
 41        0.0005571767687797546,
 42        0.0005455300561152399
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.0005829526708112098,
 49    "std": 0.00020220543356011604,
 50    "per_seed": [
 51      0.000502792012412101,
 52      0.0004844724608119577,
 53      0.0005921015981584787,
 54      0.0010409810347482562,
 55      0.000540185603313148,
 56      0.00027245740056969225,
 57      0.000595726422034204,
 58      0.0006349048344418406
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 3.7780402635689825e-05,
 64    "idea_wins": 0,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      9.700772352516651e-06,
 68      7.877941243350506e-06,
 69      0.00012473674723878503,
 70      3.3588148653507233e-06,
 71      2.852175384759903e-05,
 72      1.2275995686650276e-07,
 73      3.854965325444937e-05,
 74      8.937477832660079e-05
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 3.818063477530496e-05,
 78    "mde_rel_pct": 7.003407363892651,
 79    "verdict": "idea worse (significant)",
 80    "system_worked": false
 81  },
 82  "idea_grid": [
 83    {
 84      "lr": 0.01,
 85      "target": 0.1,
 86      "lam": 1.0
 87    },
 88    {
 89      "lr": 0.01,
 90      "target": 0.2,
 91      "lam": 1.0
 92    },
 93    {
 94      "lr": 0.01,
 95      "target": 0.3,
 96      "lam": 1.0
 97    }
 98  ],
 99  "baseline_lr_union": [
100    0.001,
101    0.003,
102    0.01
103  ],
104  "epochs": 18,
105  "n_train": 400,
106  "structural_match": "dynamics stability/control; paired end-to-end systems",
107  "parameter_count_baseline": 13313,
108  "parameter_count_idea": 13313,
109  "mechanism_signature": {
110    "baseline": {
111      "predicted_radius_target": 0.98,
112      "observed_local_jacobian_gain_mean": 0.3045346140861511,
113      "observed_full_input_jacobian_gain_mean": 0.45282769203186035,
114      "observed_local_gain_median": 0.30631518363952637,
115      "n_samples": 32,
116      "confirmed": true
117    },
118    "idea": {
119      "predicted_radius_target": 0.98,
120      "observed_local_jacobian_gain_mean": 0.2943377196788788,
121      "observed_full_input_jacobian_gain_mean": 0.44500285387039185,
122      "observed_local_gain_median": 0.29613810777664185,
123      "n_samples": 32,
124      "confirmed": true
125    },
126    "prediction": "idea should reduce trained-model local Jacobian gain toward target 0.98",
127    "confirmed": true
128  }
129}