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      "weight_decay": 0.0001
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 0.1634710617363453
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "weight_decay": 0.0001
 24        },
 25        "mean": 0.16360631212592125
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.003,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 0.02722445083782077
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "weight_decay": 0.0001
 38        },
 39        "mean": 0.027167017105966806
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.01,
 44          "weight_decay": 0.0
 45        },
 46        "mean": 0.013583156396634877
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.01,
 51          "weight_decay": 0.0001
 52        },
 53        "mean": 0.013508650125004351
 54      }
 55    ],
 56    "full": {
 57      "mean": 0.010331348807085305,
 58      "std": 0.006469534881760542,
 59      "per_seed": [
 60        0.006849721539765596,
 61        0.014833156950771809,
 62        0.0069954125210642815,
 63        0.025356309488415718,
 64        0.009617128409445286,
 65        0.0034944582730531693,
 66        0.009336123242974281,
 67        0.006168480031192303
 68      ],
 69      "n": 8
 70    }
 71  },
 72  "idea": {
 73    "mean": 0.013433240383164957,
 74    "std": 0.006230608825055202,
 75    "per_seed": [
 76      0.024367019534111023,
 77      0.015733294188976288,
 78      0.013501754030585289,
 79      0.01728116162121296,
 80      0.016352977603673935,
 81      0.009692054241895676,
 82      0.00801459327340126,
 83      0.0025230685714632273
 84    ],
 85    "n": 8,
 86    "sweep": [
 87      {
 88        "cfg": {
 89          "lr": 0.001,
 90          "lam": 0.0001
 91        },
 92        "mean": 0.16453255899250507
 93      },
 94      {
 95        "cfg": {
 96          "lr": 0.003,
 97          "lam": 0.0001
 98        },
 99        "mean": 0.029194727540016174
100      },
101      {
102        "cfg": {
103          "lr": 0.01,
104          "lam": 0.0001
105        },
106        "mean": 0.01772080734372139
107      }
108    ],
109    "best_cfg": {
110      "lr": 0.01,
111      "lam": 0.0001
112    }
113  },
114  "comparison": {
115    "delta_mean": 0.0031018915760796517,
116    "idea_wins": 3,
117    "n_pairs": 8,
118    "per_seed_diffs": [
119      0.017517297994345427,
120      0.0009001372382044792,
121      0.0065063415095210075,
122      -0.008075147867202759,
123      0.006735849194228649,
124      0.006197595968842506,
125      -0.001321529969573021,
126      -0.0036454114597290754
127    ],
128    "p_value": 0.31845,
129    "mde": 0.006606975042569378,
130    "mde_rel_pct": 63.950749954722966,
131    "verdict": "no significant win",
132    "system_worked": false
133  },
134  "mechanism_signature": {
135    "observed_increment_mse": 0.0016462175408378243,
136    "predicted_increment_mse": 0.010509551968425512,
137    "endpoint_test_mse": 0.013433240383164957,
138    "prediction": "integral residual uses observed increments instead of noisy finite-difference derivatives",
139    "confirmed": false
140  }
141}