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": 0.001
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 211.45939254760742
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 211.45939254760742
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 211.45939254760742
 29      }
 30    ],
 31    "full": {
 32      "mean": 210.28362846374512,
 33      "std": 5.264942297259805,
 34      "per_seed": [
 35        216.2948760986328,
 36        210.3721466064453,
 37        217.16351318359375,
 38        202.0070343017578,
 39        208.42745971679688,
 40        212.19456481933594,
 41        202.8369903564453,
 42        212.97244262695312
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 210.9054012298584,
 49    "std": 5.309931146588209,
 50    "per_seed": [
 51      216.95388793945312,
 52      210.90757751464844,
 53      217.41943359375,
 54      202.13302612304688,
 55      210.2574005126953,
 56      212.32080078125,
 57      203.2976531982422,
 58      213.95343017578125
 59    ],
 60    "n": 8,
 61    "config": {
 62      "lr": 0.001
 63    }
 64  },
 65  "comparison": {
 66    "delta_mean": 0.6217727661132812,
 67    "idea_wins": 0,
 68    "n_pairs": 8,
 69    "per_seed_diffs": [
 70      0.6590118408203125,
 71      0.535430908203125,
 72      0.25592041015625,
 73      0.1259918212890625,
 74      1.8299407958984375,
 75      0.1262359619140625,
 76      0.460662841796875,
 77      0.980987548828125
 78    ],
 79    "p_value": 0.0081,
 80    "mde": 0.47391836886435124,
 81    "mde_rel_pct": 0.2253710249944918,
 82    "verdict": "idea worse (significant)",
 83    "system_worked": false
 84  },
 85  "mechanism_signature": {
 86    "predicted_transition_ess": 4.0,
 87    "observed_ess_at_closest_hc": 4.193504810333252,
 88    "predicted_high_entropy_rate": 0.05,
 89    "observed_high_entropy_rate": 0.049303146592420355,
 90    "predicted_low_entropy_rate": 0.001,
 91    "observed_low_entropy_rate": 0.0011907811681947413,
 92    "mean_entropy": 0.571972486252586,
 93    "confirmed": true
 94  },
 95  "idea_nearby_settings": {
 96    "0.001": [
 97      216.95388793945312,
 98      210.90757751464844,
 99      217.41943359375,
100      202.13302612304688,
101      210.2574005126953,
102      212.32080078125,
103      203.2976531982422,
104      213.95343017578125
105    ],
106    "0.003": [
107      216.95388793945312,
108      210.90757751464844,
109      217.41943359375,
110      202.13302612304688,
111      210.2574005126953,
112      212.32080078125,
113      203.2976531982422,
114      213.95343017578125
115    ],
116    "0.01": [
117      216.95388793945312,
118      210.90757751464844,
119      217.41943359375,
120      202.13302612304688,
121      210.2574005126953,
122      212.32080078125,
123      203.2976531982422,
124      213.95343017578125
125    ]
126  },
127  "protocol_notes": "Optimizer ideas are matched to tabular Friedman regression; both systems use identical mlp_tiny, population, perturbation scale, steps, data, and paired seeds. Baseline fixed exponential cooling was swept over the shared lr union.",
128  "custom_track": null
129}