1{
  2  "bench_version": 1,
  3  "track": "torus_cayley_graph",
  4  "model": "graph_gcn_vs_relative",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.003
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.003
 15        },
 16        "mean": 2.9559298753738403
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.01
 21        },
 22        "mean": 2.957487165927887
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.03
 27        },
 28        "mean": 2.9612218737602234
 29      }
 30    ],
 31    "full": {
 32      "mean": 2.97136652469635,
 33      "std": 0.12454350264904765,
 34      "per_seed": [
 35        2.927370071411133,
 36        2.830188035964966,
 37        3.2255756855010986,
 38        2.840585947036743,
 39        3.015153169631958,
 40        2.9980242252349854,
 41        3.061436891555786,
 42        2.872598171234131
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.002413061651168391,
 49    "std": 0.0009268086641061233,
 50    "per_seed": [
 51      0.0015994536224752665,
 52      0.0017226150957867503,
 53      0.002087869681417942,
 54      0.0018045055912807584,
 55      0.0025380728766322136,
 56      0.002770226914435625,
 57      0.004656101576983929,
 58      0.0021256478503346443
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": -2.9689534630451817,
 64    "idea_wins": 8,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      -2.9257706177886575,
 68      -2.828465420869179,
 69      -3.2234878158196807,
 70      -2.8387814414454624,
 71      -3.012615096755326,
 72      -2.9952539983205497,
 73      -3.056780789978802,
 74      -2.870472523383796
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 0.11099329110372932,
 78    "mde_rel_pct": 3.73542914282754,
 79    "verdict": "idea better (significant)",
 80    "system_worked": true
 81  },
 82  "custom_track": {
 83    "name": "torus_cayley_graph",
 84    "file": "graph_track.py",
 85    "domain": "graph-nn"
 86  },
 87  "idea_lr_candidates": [
 88    {
 89      "lr": 0.003,
 90      "mean": 0.037418782245367765,
 91      "per_seed": [
 92        0.03436176851391792,
 93        0.018733929842710495,
 94        0.055476997047662735,
 95        0.040908731520175934,
 96        0.05387932434678078,
 97        0.03398267179727554,
 98        0.038339413702487946,
 99        0.02366742119193077
100      ]
101    },
102    {
103      "lr": 0.01,
104      "mean": 0.004221850220346823,
105      "per_seed": [
106        0.005596912000328302,
107        0.003076694905757904,
108        0.0050314078107476234,
109        0.0033882788848131895,
110        0.0050455559976398945,
111        0.004197766538709402,
112        0.003719017840921879,
113        0.003719167783856392
114      ]
115    },
116    {
117      "lr": 0.03,
118      "mean": 0.002413061651168391,
119      "per_seed": [
120        0.0015994536224752665,
121        0.0017226150957867503,
122        0.002087869681417942,
123        0.0018045055912807584,
124        0.0025380728766322136,
125        0.002770226914435625,
126        0.004656101576983929,
127        0.0021256478503346443
128      ]
129    }
130  ],
131  "protocol": {
132    "epochs": 18,
133    "batch": 64,
134    "paired_seeds": [
135      0,
136      1,
137      2,
138      3,
139      4,
140      5,
141      6,
142      7
143    ],
144    "structural_match": "Cayley torus graph"
145  },
146  "mechanism_signature": {
147    "baseline": {
148      "observed_directional_input_std": [
149        5.888908030792329,
150        5.888908030792329,
151        5.888908030792329,
152        5.888908030792329
153      ],
154      "predicted_sensitivity_common": -0.010245620358298103,
155      "test_mse": 2.929424524307251
156    },
157    "idea": {
158      "observed_directional_input_std": [
159        5.888908030792329,
160        5.888908030792329,
161        5.888908030792329,
162        5.888908030792329
163      ],
164      "predicted_sensitivity_common": -0.010788307526496725,
165      "test_mse": 0.005596912000328302
166    },
167    "prediction": "relation-tied message passing should exploit directional structure",
168    "confirmed": true
169  }
170}