Power-Preserving Formation GNN / bench_report.json

Failed on benchmark

Raw ⬇ ZIP
  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": 0.009
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.001250935427378863
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.0007842454651836306
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.009
 27        },
 28        "mean": 0.00032731111787143163
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.0002811976373777725,
 33      "std": 0.00014349708189490275,
 34      "per_seed": [
 35        0.0002109479537466541,
 36        0.00023838288325350732,
 37        0.00020579870033543557,
 38        0.0006541149341501296,
 39        0.0002111889043590054,
 40        0.00019325825269334018,
 41        0.00025733737857080996,
 42        0.00027855209191329777
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "best_cfg": {
 49      "lr": 0.009
 50    },
 51    "sweep": [
 52      {
 53        "cfg": {
 54          "lr": 0.001
 55        },
 56        "mean": 0.15046435873955488
 57      },
 58      {
 59        "cfg": {
 60          "lr": 0.003
 61        },
 62        "mean": 0.009112948318943381
 63      },
 64      {
 65        "cfg": {
 66          "lr": 0.009
 67        },
 68        "mean": 0.002690640962100588
 69      }
 70    ],
 71    "per_seed": [
 72      0.002644516061991453,
 73      0.0025255768559873104,
 74      0.0035836552269756794,
 75      0.003749588970094919,
 76      0.0022452210541814566,
 77      0.0017906847642734647,
 78      0.0021657501347362995,
 79      0.0028201346285641193
 80    ],
 81    "mean": 0.002690640962100588,
 82    "std": 0.0006372615308025123,
 83    "n": 8
 84  },
 85  "comparison": {
 86    "delta_mean": 0.0024094433247228153,
 87    "idea_wins": 0,
 88    "n_pairs": 8,
 89    "per_seed_diffs": [
 90      0.002433568108244799,
 91      0.002287193972733803,
 92      0.003377856526640244,
 93      0.0030954740359447896,
 94      0.002034032149822451,
 95      0.0015974265115801245,
 96      0.0019084127561654896,
 97      0.0025415825366508216
 98    ],
 99    "p_value": 0.0081,
100    "mde": 0.0004989760120697837,
101    "mde_rel_pct": 177.44672989532938,
102    "verdict": "idea worse (significant)",
103    "system_worked": false
104  },
105  "track_justification": "Dynamics is structurally matched: the idea claims stability/control of repeated state propagation.",
106  "mechanism_signature": {
107    "energy_first_mean": 0.0,
108    "energy_last_mean": 0.21861755847930908,
109    "max_step_energy_increase": 0.05179046094417572,
110    "mean_continuous_power": 0.47418127581477165,
111    "prediction": "bounded hidden energy under damped port-Hamiltonian updates",
112    "confirmed": false
113  }
114}