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": 0.02100143244024366
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.0030831422773189843
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.002158296701963991
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.0020533509959932417,
 33      "std": 0.0007424383658550215,
 34      "per_seed": [
 35        0.002525518648326397,
 36        0.0009679240174591541,
 37        0.0023394303862005472,
 38        0.0028003137558698654,
 39        0.0015804190188646317,
 40        0.003060286631807685,
 41        0.002158003393560648,
 42        0.0009949121158570051
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.07345113484188914,
 49    "std": 0.011031271199206346,
 50    "per_seed": [
 51      0.06164592131972313,
 52      0.07005804777145386,
 53      0.08214986324310303,
 54      0.09056025743484497,
 55      0.05593807250261307,
 56      0.07245660573244095,
 57      0.08499487489461899,
 58      0.06980543583631516
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.0713977838458959,
 64    "idea_wins": 0,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      0.05912040267139673,
 68      0.0690901237539947,
 69      0.07981043285690248,
 70      0.0877599436789751,
 71      0.054357653483748436,
 72      0.06939631910063326,
 73      0.08283687150105834,
 74      0.06881052372045815
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 0.009619511434093848,
 78    "mde_rel_pct": 468.4786698847229,
 79    "verdict": "idea worse (significant)",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "mechanism_signature": {
 84      "predicted": "trajectory velocity should vary with segment/time and integration should reconstruct endpoint",
 85      "baseline_endpoint_mse": 0.003087950637564063,
 86      "idea_endpoint_mse": 0.05978367105126381,
 87      "initial_velocity_mse": 0.04817172512412071,
 88      "confirmed": true
 89    },
 90    "idea_sweep": [
 91      {
 92        "cfg": {
 93          "lr": 0.01
 94        },
 95        "mean": 0.07610352244228125
 96      },
 97      {
 98        "cfg": {
 99          "lr": 0.001
100        },
101        "mean": 0.10371190309524536
102      },
103      {
104        "cfg": {
105          "lr": 0.01
106        },
107        "mean": 0.07610352244228125
108      }
109    ],
110    "track_match": "dynamics matches controlled solver/refinement trajectories"
111  },
112  "math_check": {
113    "interpolation_identity_max_error": 0.0,
114    "velocity_identity_max_error": 4.678964614868164e-06,
115    "prediction": "piecewise interpolation and segment velocity are exact"
116  },
117  "baseline_union_grid": [
118    {
119      "lr": 0.001
120    },
121    {
122      "lr": 0.003
123    },
124    {
125      "lr": 0.01
126    }
127  ],
128  "idea_best_cfg": {
129    "lr": 0.01
130  }
131}