1{
  2  "bench_version": 1,
  3  "track": "cycle_parity_graph",
  4  "model": "graphnet",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.003,
 10      "epochs": 12
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "epochs": 12
 17        },
 18        "mean": 0.3074999889358878
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "epochs": 12
 24        },
 25        "mean": 0.2012499978300184
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.006,
 30          "epochs": 12
 31        },
 32        "mean": 0.2162499960977584
 33      }
 34    ],
 35    "full": {
 36      "mean": 0.20874999731313437,
 37      "std": 0.20704392538301097,
 38      "per_seed": [
 39        0.0,
 40        0.28999999165534973,
 41        0.5,
 42        0.014999999664723873,
 43        0.5,
 44        0.32499998807907104,
 45        0.0,
 46        0.03999999910593033
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "mean": 0.0,
 53    "std": 0.0,
 54    "per_seed": [
 55      0.0,
 56      0.0,
 57      0.0,
 58      0.0,
 59      0.0,
 60      0.0,
 61      0.0,
 62      0.0
 63    ],
 64    "n": 8
 65  },
 66  "comparison": {
 67    "delta_mean": -0.20874999731313437,
 68    "idea_wins": 6,
 69    "n_pairs": 8,
 70    "per_seed_diffs": [
 71      0.0,
 72      -0.28999999165534973,
 73      -0.5,
 74      -0.014999999664723873,
 75      -0.5,
 76      -0.32499998807907104,
 77      0.0,
 78      -0.03999999910593033
 79    ],
 80    "p_value": 0.0312,
 81    "mde": 0.18507366186801333,
 82    "mde_rel_pct": 88.65804275455608,
 83    "verdict": "idea better (significant)",
 84    "system_worked": true
 85  },
 86  "mechanism_signature": {
 87    "math_checks": {
 88      "odd_degree_max_residual": 0.0,
 89      "parity": [
 90        {
 91          "edges": 1,
 92          "max_error": 0
 93        },
 94        {
 95          "edges": 2,
 96          "max_error": 0
 97        },
 98        {
 99          "edges": 3,
100          "max_error": 0
101        },
102        {
103          "edges": 4,
104          "max_error": 0
105        },
106        {
107          "edges": 5,
108          "max_error": 0
109        },
110        {
111          "edges": 6,
112          "max_error": 0
113        },
114        {
115          "edges": 7,
116          "max_error": 0
117        },
118        {
119          "edges": 8,
120          "max_error": 0
121        }
122      ],
123      "wedge_energy": {
124        "predicted": 24,
125        "observed": 23.964393615722656,
126        "ratio": 0.9985164006551107
127      }
128    },
129    "prediction": "odd incident degree is annihilated exactly at NN scale",
130    "predicted_max_residual": 0.0,
131    "observed_max_residual": 0.0,
132    "confirmed": true,
133    "n_observations": 24
134  },
135  "custom_track": {
136    "name": "cycle_parity_graph",
137    "file": "bench_fermionic.py",
138    "domain": "graph-nn"
139  }
140}