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": 0.01,
 10      "weight_decay": 0.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 52.02932357788086
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "weight_decay": 0.0
 24        },
 25        "mean": 14.625996589660645
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.01,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 9.001914024353027
 33      }
 34    ],
 35    "full": {
 36      "mean": 9.259196519851685,
 37      "std": 1.0373002651087238,
 38      "per_seed": [
 39        8.302460670471191,
 40        10.292799949645996,
 41        8.400059700012207,
 42        9.012335777282715,
 43        7.509613037109375,
 44        9.842769622802734,
 45        10.44126033782959,
 46        10.272273063659668
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "mean": 15.81218934059143,
 53    "std": 1.949355068505579,
 54    "per_seed": [
 55      16.183080673217773,
 56      15.5823335647583,
 57      17.167373657226562,
 58      13.497793197631836,
 59      12.077912330627441,
 60      16.422466278076172,
 61      17.101774215698242,
 62      18.464780807495117
 63    ],
 64    "n": 8,
 65    "best_cfg": {
 66      "lr": 0.003,
 67      "weight_decay": 0.0
 68    }
 69  },
 70  "comparison": {
 71    "delta_mean": 6.552992820739746,
 72    "idea_wins": 0,
 73    "n_pairs": 8,
 74    "per_seed_diffs": [
 75      7.880620002746582,
 76      5.289533615112305,
 77      8.767313957214355,
 78      4.485457420349121,
 79      4.568299293518066,
 80      6.5796966552734375,
 81      6.660513877868652,
 82      8.19250774383545
 83    ],
 84    "p_value": 0.0081,
 85    "mde": 1.3839413130321554,
 86    "mde_rel_pct": 14.94666745721392,
 87    "verdict": "idea worse (significant)",
 88    "system_worked": false
 89  },
 90  "track_choice": "tabular is the built-in optimizer/training-dynamics match; Friedman regression keeps architecture and task fixed.",
 91  "idea_sweep": [
 92    {
 93      "cfg": {
 94        "lr": 0.001,
 95        "weight_decay": 0.0
 96      },
 97      "result": {
 98        "mean": 76.82785987854004,
 99        "std": 25.70975037882407,
100        "per_seed": [
101          70.13491821289062,
102          102.93740844726562,
103          84.74547576904297,
104          91.09562683105469,
105          33.59514236450195,
106          98.26583099365234,
107          37.17234420776367,
108          96.67613220214844
109        ],
110        "n": 8
111      }
112    },
113    {
114      "cfg": {
115        "lr": 0.003,
116        "weight_decay": 0.0
117      },
118      "result": {
119        "mean": 15.81218934059143,
120        "std": 1.949355068505579,
121        "per_seed": [
122          16.183080673217773,
123          15.5823335647583,
124          17.167373657226562,
125          13.497793197631836,
126          12.077912330627441,
127          16.422466278076172,
128          17.101774215698242,
129          18.464780807495117
130        ],
131        "n": 8,
132        "best_cfg": {
133          "lr": 0.003,
134          "weight_decay": 0.0
135        }
136      }
137    },
138    {
139      "cfg": {
140        "lr": 0.01,
141        "weight_decay": 0.0
142      },
143      "result": {
144        "mean": 34.88474154472351,
145        "std": 15.230981751784949,
146        "per_seed": [
147          34.66309356689453,
148          59.120567321777344,
149          43.088172912597656,
150          53.1762580871582,
151          21.203292846679688,
152          34.56721496582031,
153          14.101446151733398,
154          19.157886505126953
155        ],
156        "n": 8
157      }
158    }
159  ],
160  "budget": {
161    "baseline_epochs": 16,
162    "idea_generations": 16,
163    "population": 8
164  },
165  "mechanism_signature": {
166    "observed_cov_increment": 0.000783836052484821,
167    "predicted_2D_increment": 0.0005999999999999998,
168    "relative_error": 0.3063934202973796,
169    "curvature_median_final": -8.0,
170    "confirmed": true,
171    "trained_model_metric": 16.183080673217773
172  }
173}