1{
  2  "bench_version": 1,
  3  "track": "poisson_boundary",
  4  "model": "mlp_tiny",
  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": 0.06875268183648586
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.05776503775268793
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.04322445672005415
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.043870826717466116,
 33      "std": 0.002839556331423274,
 34      "per_seed": [
 35        0.04028090834617615,
 36        0.044616829603910446,
 37        0.04462457448244095,
 38        0.043375514447689056,
 39        0.038494233042001724,
 40        0.04666280001401901,
 41        0.046755287796258926,
 42        0.046156466007232666
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.06319879228249192,
 49    "std": 0.011146065364718148,
 50    "per_seed": [
 51      0.04028090834617615,
 52      0.07205105572938919,
 53      0.07206139713525772,
 54      0.055017322301864624,
 55      0.05459468439221382,
 56      0.07074009627103806,
 57      0.07260646671056747,
 58      0.06823840737342834
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.019327965565025806,
 64    "idea_wins": 0,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      0.0,
 68      0.027434226125478745,
 69      0.027436822652816772,
 70      0.011641807854175568,
 71      0.016100451350212097,
 72      0.024077296257019043,
 73      0.025851178914308548,
 74      0.02208194136619568
 75    ],
 76    "p_value": 0.01485,
 77    "mde": 0.008040142777977925,
 78    "mde_rel_pct": 18.326854950232647,
 79    "verdict": "idea worse (significant)",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "mechanism_signature": {
 84      "selected_index": 5,
 85      "scores": [
 86        3.491577856996373,
 87        3.515875502902411,
 88        3.537788184248112,
 89        3.5629319410966094,
 90        3.474729111946311,
 91        3.402205193851991
 92      ],
 93      "nested_probe": {
 94        "2": 3.402205193851991,
 95        "3": 3.536084817812452,
 96        "4": 3.5380539738679375
 97      },
 98      "predicted_monotone": true,
 99      "confirmed": true
100    },
101    "custom_track": {
102      "name": "poisson_boundary",
103      "file": "poisson_track.py",
104      "domain": "pde"
105    }
106  },
107  "idea_config_sweep": [
108    {
109      "cfg": {
110        "lr": 0.01,
111        "level": 2
112      },
113      "mean": 0.06319879228249192,
114      "per_seed": [
115        0.04028090834617615,
116        0.07205105572938919,
117        0.07206139713525772,
118        0.055017322301864624,
119        0.05459468439221382,
120        0.07074009627103806,
121        0.07260646671056747,
122        0.06823840737342834
123      ]
124    },
125    {
126      "cfg": {
127        "lr": 0.01,
128        "level": 3
129      },
130      "mean": 0.06653260113671422,
131      "per_seed": [
132        0.06695137917995453,
133        0.07205105572938919,
134        0.07206139713525772,
135        0.055017322301864624,
136        0.05459468439221382,
137        0.07074009627103806,
138        0.07260646671056747,
139        0.06823840737342834
140      ]
141    },
142    {
143      "cfg": {
144        "lr": 0.01,
145        "level": 4
146      },
147      "mean": 0.06653260113671422,
148      "per_seed": [
149        0.06695137917995453,
150        0.07205105572938919,
151        0.07206139713525772,
152        0.055017322301864624,
153        0.05459468439221382,
154        0.07074009627103806,
155        0.07260646671056747,
156        0.06823840737342834
157      ]
158    }
159  ],
160  "runtime_seconds": 117.34046530723572,
161  "protocol_note": "Custom PDE track required; baseline and idea share mlp_tiny and Adam, differing only in archived checkpoint selector."
162}