1{
  2  "bench_version": 1,
  3  "track": "tabular",
  4  "model": "mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.008
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 61.77081108093262
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.002
 21        },
 22        "mean": 17.307271003723145
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.003
 27        },
 28        "mean": 15.343324422836304
 29      },
 30      {
 31        "cfg": {
 32          "lr": 0.005
 33        },
 34        "mean": 12.93674373626709
 35      },
 36      {
 37        "cfg": {
 38          "lr": 0.008
 39        },
 40        "mean": 10.50437068939209
 41      }
 42    ],
 43    "full": {
 44      "mean": 10.8468017578125,
 45      "std": 0.9460715154353679,
 46      "per_seed": [
 47        10.433314323425293,
 48        9.383343696594238,
 49        12.219518661499023,
 50        9.981306076049805,
 51        10.63367748260498,
 52        10.9376859664917,
 53        10.86903190612793,
 54        12.316535949707031
 55      ],
 56      "n": 8
 57    }
 58  },
 59  "idea": {
 60    "mean": 8.208642661571503,
 61    "std": 0.5742573828718772,
 62    "per_seed": [
 63      7.412168025970459,
 64      7.734030246734619,
 65      8.057981491088867,
 66      7.578333377838135,
 67      8.25620174407959,
 68      8.907169342041016,
 69      8.773388862609863,
 70      8.949868202209473
 71    ],
 72    "n": 8
 73  },
 74  "comparison": {
 75    "delta_mean": -2.6381590962409973,
 76    "idea_wins": 8,
 77    "n_pairs": 8,
 78    "per_seed_diffs": [
 79      -3.021146297454834,
 80      -1.6493134498596191,
 81      -4.161537170410156,
 82      -2.40297269821167,
 83      -2.3774757385253906,
 84      -2.0305166244506836,
 85      -2.0956430435180664,
 86      -3.3666677474975586
 87    ],
 88    "p_value": 0.0081,
 89    "mde": 0.6897071822966465,
 90    "mde_rel_pct": 6.3586225478849485,
 91    "verdict": "idea better (significant)",
 92    "system_worked": true
 93  },
 94  "track_choice": "tabular: the idea is an optimizer/black-box hyperparameter infill method; no architecture or sequence/control structure is required.",
 95  "shared_lr_union": [
 96    0.001,
 97    0.002,
 98    0.003,
 99    0.005,
100    0.008
101  ],
102  "epochs": 12,
103  "n_train": 800,
104  "mechanism_signature": {
105    "pilot_validation_mean": 7.560150101780891,
106    "gn_predicted_validation_mean": 8.995661415256667,
107    "gn_observed_validation_mean": 8.700061082839966,
108    "absolute_prediction_error": 0.5079499775828902,
109    "condition_estimate": 1.0,
110    "confirmed": true
111  },
112  "math_check": {
113    "exact_linear_jacobian_relative_error": 0.8606629657653821
114  }
115}