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": 0.006,
 10      "epochs": 18
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.0015,
 16          "epochs": 18
 17        },
 18        "mean": 16.413997888565063
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "epochs": 18
 24        },
 25        "mean": 13.256237506866455
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.006,
 30          "epochs": 18
 31        },
 32        "mean": 10.138120174407959
 33      }
 34    ],
 35    "full": {
 36      "mean": 10.593586206436157,
 37      "std": 1.1860413365771312,
 38      "per_seed": [
 39        9.266413688659668,
 40        10.942156791687012,
 41        10.532804489135742,
 42        9.811105728149414,
 43        9.271466255187988,
 44        10.483624458312988,
 45        11.298463821411133,
 46        13.142654418945312
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "mean": 24.50470805168152,
 53    "std": 1.4738344415396776,
 54    "per_seed": [
 55      25.35301971435547,
 56      24.542402267456055,
 57      26.259077072143555,
 58      22.042612075805664,
 59      22.800880432128906,
 60      25.55665397644043,
 61      23.438636779785156,
 62      26.044382095336914
 63    ],
 64    "n": 8,
 65    "best_cfg": {
 66      "lr": 0.003,
 67      "epochs": 18
 68    },
 69    "settings": [
 70      {
 71        "cfg": {
 72          "lr": 0.0015,
 73          "epochs": 18
 74        },
 75        "mean": 48.017033100128174
 76      },
 77      {
 78        "cfg": {
 79          "lr": 0.003,
 80          "epochs": 18
 81        },
 82        "mean": 24.50470805168152
 83      },
 84      {
 85        "cfg": {
 86          "lr": 0.006,
 87          "epochs": 18
 88        },
 89        "mean": 24.52953600883484
 90      }
 91    ]
 92  },
 93  "comparison": {
 94    "delta_mean": 13.911121845245361,
 95    "idea_wins": 0,
 96    "n_pairs": 8,
 97    "per_seed_diffs": [
 98      16.0866060256958,
 99      13.600245475769043,
100      15.726272583007812,
101      12.23150634765625,
102      13.529414176940918,
103      15.073029518127441,
104      12.140172958374023,
105      12.901727676391602
106    ],
107    "p_value": 0.0081,
108    "mde": 1.2874798773006364,
109    "mde_rel_pct": 12.153390289290563,
110    "verdict": "idea worse (significant)",
111    "system_worked": false
112  },
113  "track_choice": "Friedman#1 is the built-in regression track and matches the idea's high-dimensional regression setting.",
114  "custom_track": null,
115  "parameter_counts": {
116    "baseline": 4929,
117    "idea": 7076
118  },
119  "mechanism_signature": {
120    "definition": "trained accumulator behavior on tabular test predictions",
121    "predicted": "context is transmitted unchanged and output corrections accumulate additively",
122    "observed_mean_delta_norms": [
123      0.9999635517597198,
124      0.9999598488211632
125    ],
126    "observed_final_accumulator_norm": 2.128704696893692,
127    "observed_prediction_std": 0.06401639571413398,
128    "quantitative_tolerance": "nonzero branch corrections and finite accumulated state",
129    "confirmed": true
130  }
131}