1{
  2  "bench_version": 1,
  3  "track": "minkowski_composition",
  4  "model": "mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.01,
 10      "weight_decay": 0.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 0.11922426894307137
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "weight_decay": 0.0
 24        },
 25        "mean": 0.057413769885897636
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.01,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 0.022922257892787457
 33      }
 34    ],
 35    "full": {
 36      "mean": 0.020459107705391943,
 37      "std": 0.0058978451511714595,
 38      "per_seed": [
 39        0.02743070386350155,
 40        0.02446400746703148,
 41        0.017893968150019646,
 42        0.021900352090597153,
 43        0.011625968851149082,
 44        0.025349821895360947,
 45        0.010963048785924911,
 46        0.02404499053955078
 47      ],
 48      "n": 8
 49    },
 50    "idea_union_grid": [
 51      {
 52        "lr": 0.001,
 53        "weight_decay": 0.0
 54      },
 55      {
 56        "lr": 0.003,
 57        "weight_decay": 0.0
 58      },
 59      {
 60        "lr": 0.01,
 61        "weight_decay": 0.0
 62      }
 63    ]
 64  },
 65  "idea": {
 66    "mean": 0.0010090162832057104,
 67    "std": 0.0002896170011232185,
 68    "per_seed": [
 69      0.00107665010727942,
 70      0.0011185563635081053,
 71      0.0006614169105887413,
 72      0.0014436655910685658,
 73      0.0006137294112704694,
 74      0.0012164852814748883,
 75      0.0007038944750092924,
 76      0.0012377321254462004
 77    ],
 78    "n": 8
 79  },
 80  "comparison": {
 81    "delta_mean": -0.019450091422186233,
 82    "idea_wins": 8,
 83    "n_pairs": 8,
 84    "per_seed_diffs": [
 85      -0.02635405375622213,
 86      -0.023345451103523374,
 87      -0.017232551239430904,
 88      -0.020456686499528587,
 89      -0.011012239439878613,
 90      -0.02413333661388606,
 91      -0.01025915431091562,
 92      -0.02280725841410458
 93    ],
 94    "p_value": 0.0081,
 95    "mde": 0.005072785876795095,
 96    "mde_rel_pct": 24.79475620267728,
 97    "verdict": "idea better (significant)",
 98    "system_worked": true
 99  },
100  "custom_track": {
101    "name": "minkowski_composition",
102    "file": "minkowski_track.py",
103    "domain": "structured_convex_composition"
104  },
105  "idea_sweep": [
106    {
107      "cfg": {
108        "lr": 0.001,
109        "weight_decay": 0.0
110      },
111      "mean": 0.002880843123421073
112    },
113    {
114      "cfg": {
115        "lr": 0.003,
116        "weight_decay": 0.0
117      },
118      "mean": 0.0020782598585356027
119    },
120    {
121      "cfg": {
122        "lr": 0.01,
123        "weight_decay": 0.0
124      },
125      "mean": 0.001075072243111208
126    }
127  ],
128  "mechanism_signature": {
129    "quantity": "trained additive output error and local perturbation amplification",
130    "predicted": {
131      "componentwise_addition": "nonnegative weighted sum",
132      "amplification_bound": "approximately alpha+beta"
133    },
134    "observed_mean_additivity_error": 0.10632766596972942,
135    "observed_median_local_amplification": 1.3048688173294067,
136    "observed_p90_local_amplification": 1.5228296875953675,
137    "confirmed": true
138  }
139}