1{
  2  "bench_version": 1,
  3  "track": "planar_ising_autoregressive",
  4  "model": "mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.01,
 10      "epochs": 5,
 11      "batch": 128
 12    },
 13    "sweep": [
 14      {
 15        "cfg": {
 16          "lr": 0.001,
 17          "epochs": 5,
 18          "batch": 128
 19        },
 20        "mean": 0.1614346206188202
 21      },
 22      {
 23        "cfg": {
 24          "lr": 0.003,
 25          "epochs": 5,
 26          "batch": 128
 27        },
 28        "mean": 0.13007085770368576
 29      },
 30      {
 31        "cfg": {
 32          "lr": 0.01,
 33          "epochs": 5,
 34          "batch": 128
 35        },
 36        "mean": 0.12258148938417435
 37      }
 38    ],
 39    "full": {
 40      "mean": 0.11785224732011557,
 41      "std": 0.017848149462034908,
 42      "per_seed": [
 43        0.15073059499263763,
 44        0.13608941435813904,
 45        0.11354892700910568,
 46        0.08995702117681503,
 47        0.11521363258361816,
 48        0.10482487082481384,
 49        0.12506559491157532,
 50        0.10738792270421982
 51      ],
 52      "n": 8
 53    }
 54  },
 55  "idea": {
 56    "mean": 0.11573431827127934,
 57    "std": 0.01748951353530187,
 58    "per_seed": [
 59      0.14706209301948547,
 60      0.1326635181903839,
 61      0.11457636952400208,
 62      0.09001504629850388,
 63      0.10833320766687393,
 64      0.09629029035568237,
 65      0.12431652843952179,
 66      0.11261749267578125
 67    ],
 68    "n": 8
 69  },
 70  "comparison": {
 71    "delta_mean": -0.0021179290488362312,
 72    "idea_wins": 5,
 73    "n_pairs": 8,
 74    "per_seed_diffs": [
 75      -0.0036685019731521606,
 76      -0.003425896167755127,
 77      0.0010274425148963928,
 78      5.8025121688842773e-05,
 79      -0.006880424916744232,
 80      -0.00853458046913147,
 81      -0.0007490664720535278,
 82      0.005229569971561432
 83    ],
 84    "p_value": 0.23005,
 85    "mde": 0.003718736254444294,
 86    "mde_rel_pct": 3.1554224369971453,
 87    "verdict": "no significant win",
 88    "system_worked": false
 89  },
 90  "mechanism_signature": {
 91    "prediction": "exact soft conditional labels reduce trained-NN conditional probability MSE versus sampled labels",
 92    "baseline_seed0": {
 93      "metric": 0.15073059499263763,
 94      "q_mse": 0.03577027469873428,
 95      "pred_mean": 0.4622843861579895
 96    },
 97    "idea_seed0": {
 98      "metric": 0.14706209301948547,
 99      "q_mse": 0.03235262259840965,
100      "pred_mean": 0.4577365219593048
101    },
102    "q_mse_reduction": 0.0034176521003246307,
103    "confirmed": true
104  },
105  "idea_sweep": [
106    {
107      "cfg": {
108        "lr": 0.001,
109        "epochs": 5,
110        "batch": 128
111      },
112      "result": {
113        "mean": 0.17583313025534153,
114        "std": 0.04598349937323497,
115        "per_seed": [
116          0.15006619691848755,
117          0.1609499603509903,
118          0.18380002677440643,
119          0.15149202942848206,
120          0.15694455802440643,
121          0.2921749949455261,
122          0.17385314404964447,
123          0.13738413155078888
124        ],
125        "n": 8
126      }
127    },
128    {
129      "cfg": {
130        "lr": 0.003,
131        "epochs": 5,
132        "batch": 128
133      },
134      "result": {
135        "mean": 0.12854224257171154,
136        "std": 0.01281968850255869,
137        "per_seed": [
138          0.1471148431301117,
139          0.14259950816631317,
140          0.1184714213013649,
141          0.11075853556394577,
142          0.13586729764938354,
143          0.11389961838722229,
144          0.136264368891716,
145          0.12336234748363495
146        ],
147        "n": 8
148      }
149    },
150    {
151      "cfg": {
152        "lr": 0.01,
153        "epochs": 5,
154        "batch": 128
155      },
156      "result": {
157        "mean": 0.11573431827127934,
158        "std": 0.01748951353530187,
159        "per_seed": [
160          0.14706209301948547,
161          0.1326635181903839,
162          0.11457636952400208,
163          0.09001504629850388,
164          0.10833320766687393,
165          0.09629029035568237,
166          0.12431652843952179,
167          0.11261749267578125
168        ],
169        "n": 8
170      }
171    }
172  ],
173  "custom_track": {
174    "name": "planar_ising_autoregressive",
175    "file": "ising_teacher_track.py",
176    "domain": "sequence-level sampling"
177  }
178}