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": 0.004247799690347165
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.0018907261837739497
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.0005508567701326683
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.00043304020073264837,
 33      "std": 0.00018680572581609252,
 34      "per_seed": [
 35        0.000348469679011032,
 36        0.00044166683801449835,
 37        0.0005423855036497116,
 38        0.0008709050598554313,
 39        0.00039059933624230325,
 40        0.0003389842459000647,
 41        0.0002937567769549787,
 42        0.00023755416623316705
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 1.8605211664635135e-05,
 49    "std": 2.831933136541298e-05,
 50    "per_seed": [
 51      1.025223082251614e-05,
 52      5.093114396004239e-06,
 53      3.275605195085518e-06,
 54      2.0394263628986664e-05,
 55      5.965866421320243e-06,
 56      7.376182111329399e-06,
 57      9.230343130184337e-05,
 58      4.180999439995503e-06
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": -0.00041443498906801324,
 64    "idea_wins": 8,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      -0.00033821744818851585,
 68      -0.0004365737236184941,
 69      -0.0005391098984546261,
 70      -0.0008505107962264447,
 71      -0.000384633469820983,
 72      -0.0003316080637887353,
 73      -0.00020145334565313533,
 74      -0.00023337316679317155
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 0.0001724928317781249,
 78    "mde_rel_pct": 39.832983516608664,
 79    "verdict": "idea better (significant)",
 80    "system_worked": true
 81  },
 82  "mechanism_signature": {
 83    "predicted_slowest_memory_mode": 0.3352634608745575,
 84    "observed_max_state_norm": 2.583609104156494,
 85    "bounded_observed_state": true,
 86    "confirmed": true
 87  },
 88  "idea_sweep": [
 89    {
 90      "cfg": {
 91        "lr": 0.001
 92      },
 93      "mean": 5.264711217023432e-05
 94    },
 95    {
 96      "cfg": {
 97        "lr": 0.003
 98      },
 99      "mean": 1.748438717186218e-05
100    },
101    {
102      "cfg": {
103        "lr": 0.01
104      },
105      "mean": 9.75380351064814e-06
106    }
107  ],
108  "idea_best_cfg": {
109    "lr": 0.01
110  },
111  "track_justification": "Dynamics is structurally matched: the task is actuated pendulum forecasting and the intervention is a stable recurrent memory state.",
112  "search_space_parity": {
113    "baseline_grid": [
114      0.001,
115      0.003,
116      0.01
117    ],
118    "idea_grid": [
119      0.001,
120      0.003,
121      0.01
122    ],
123    "selected_idea_lr": 0.01
124  },
125  "parameterization": {
126    "baseline_params": 3585,
127    "idea_params": 16077,
128    "epochs": 12,
129    "n_train": 1000,
130    "n_test": 300
131  }
132}