1{
  2  "bench_version": 1,
  3  "track": "tabular",
  4  "model": "candidate_mlp_shared",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.01
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 119.02437210083008
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 30.62220811843872
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 21.151901245117188
 29      }
 30    ],
 31    "full": {
 32      "mean": 21.90489387512207,
 33      "std": 1.9880925147812747,
 34      "per_seed": [
 35        23.918453216552734,
 36        18.010822296142578,
 37        22.42278480529785,
 38        20.255544662475586,
 39        22.012264251708984,
 40        22.661293029785156,
 41        21.173133850097656,
 42        24.784854888916016
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 21.988378763198853,
 49    "std": 1.863640983270208,
 50    "per_seed": [
 51      23.71569061279297,
 52      18.270889282226562,
 53      22.492475509643555,
 54      20.57891082763672,
 55      22.177988052368164,
 56      22.6539249420166,
 57      21.250635147094727,
 58      24.766515731811523
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.08348488807678223,
 64    "idea_wins": 3,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      -0.20276260375976562,
 68      0.2600669860839844,
 69      0.06969070434570312,
 70      0.3233661651611328,
 71      0.1657238006591797,
 72      -0.0073680877685546875,
 73      0.07750129699707031,
 74      -0.018339157104492188
 75    ],
 76    "p_value": 0.18505,
 77    "mde": 0.14001291226171048,
 78    "mde_rel_pct": 0.639185531141635,
 79    "verdict": "no measurable effect",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "predicted_projected_gain_mean": 26.90758864581585,
 84    "observed_heldout_selected_response_mse_mean": 65.35227632522583,
 85    "predicted_per_seed": [
 86      21.2170729637146,
 87      7.129383563995361,
 88      23.922744750976562,
 89      49.054978370666504,
 90      8.374441027641296,
 91      25.982917308807373,
 92      46.1535587310791,
 93      33.425612449645996
 94    ],
 95    "observed_per_seed": [
 96      64.23245239257812,
 97      69.09458923339844,
 98      59.128604888916016,
 99      57.94867706298828,
100      67.77635955810547,
101      68.72184753417969,
102      67.34278106689453,
103      68.5728988647461
104    ],
105    "correlation": -0.415484140116091,
106    "confirmed": false
107  },
108  "idea_sweep": [
109    {
110      "cfg": {
111        "lr": 0.001
112      },
113      "result": {
114        "mean": 117.0982551574707,
115        "std": 6.973504377883443,
116        "per_seed": [
117          121.33436584472656,
118          115.73588562011719,
119          113.34111785888672,
120          109.39837646484375,
121          118.32662963867188,
122          115.21646118164062,
123          110.5782699584961,
124          132.8549346923828
125        ],
126        "n": 8
127      }
128    },
129    {
130      "cfg": {
131        "lr": 0.003
132      },
133      "result": {
134        "mean": 31.2304470539093,
135        "std": 3.2026968213759157,
136        "per_seed": [
137          33.56842803955078,
138          26.45305633544922,
139          31.497737884521484,
140          29.350759506225586,
141          30.496389389038086,
142          29.599700927734375,
143          30.8070011138916,
144          38.07050323486328
145        ],
146        "n": 8
147      }
148    },
149    {
150      "cfg": {
151        "lr": 0.01
152      },
153      "result": {
154        "mean": 21.988378763198853,
155        "std": 1.863640983270208,
156        "per_seed": [
157          23.71569061279297,
158          18.270889282226562,
159          22.492475509643555,
160          20.57891082763672,
161          22.177988052368164,
162          22.6539249420166,
163          21.250635147094727,
164          24.766515731811523
165        ],
166        "n": 8
167      }
168    }
169  ],
170  "selection": {
171    "candidates": 8,
172    "active_blocks": 3,
173    "epochs": 18,
174    "n_train": 400,
175    "n_test": 200,
176    "track_rationale": "tabular regression is the built-in optimizer/regularizer/architecture-compatible small MLP track; no PDE/control structure is required by this adapter-bank idea"
177  }
178}