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      "boundary_weight": 1.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "boundary_weight": 1.0
 17        },
 18        "mean": 0.010570373153313994
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "boundary_weight": 10.0
 24        },
 25        "mean": 0.014104659901931882
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.001,
 30          "boundary_weight": 30.0
 31        },
 32        "mean": 0.05715882801450789
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "boundary_weight": 1.0
 38        },
 39        "mean": 0.008062692591920495
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.003,
 44          "boundary_weight": 10.0
 45        },
 46        "mean": 0.010542358737438917
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.003,
 51          "boundary_weight": 30.0
 52        },
 53        "mean": 0.014223965117707849
 54      },
 55      {
 56        "cfg": {
 57          "lr": 0.01,
 58          "boundary_weight": 1.0
 59        },
 60        "mean": 0.007657136651687324
 61      },
 62      {
 63        "cfg": {
 64          "lr": 0.01,
 65          "boundary_weight": 10.0
 66        },
 67        "mean": 0.01242852327413857
 68      },
 69      {
 70        "cfg": {
 71          "lr": 0.01,
 72          "boundary_weight": 30.0
 73        },
 74        "mean": 0.04939912143163383
 75      }
 76    ],
 77    "full": {
 78      "mean": 0.007044873433187604,
 79      "std": 0.00100136969136606,
 80      "per_seed": [
 81        0.006205223500728607,
 82        0.00873982161283493,
 83        0.008036286570131779,
 84        0.00764721492305398,
 85        0.005511410068720579,
 86        0.006294506136327982,
 87        0.0072539327666163445,
 88        0.00667059188708663
 89      ],
 90      "n": 8
 91    }
 92  },
 93  "idea": {
 94    "mean": 0.006475340866018087,
 95    "std": 0.0010829860742993793,
 96    "per_seed": [
 97      0.005347711965441704,
 98      0.007960012182593346,
 99      0.00744650699198246,
100      0.006636394187808037,
101      0.004956180229783058,
102      0.005183100700378418,
103      0.007344032172113657,
104      0.006928788498044014
105    ],
106    "n": 8
107  },
108  "comparison": {
109    "delta_mean": -0.0005695325671695173,
110    "idea_wins": 6,
111    "n_pairs": 8,
112    "per_seed_diffs": [
113      -0.0008575115352869034,
114      -0.0007798094302415848,
115      -0.0005897795781493187,
116      -0.001010820735245943,
117      -0.000555229838937521,
118      -0.001111405435949564,
119      9.009940549731255e-05,
120      0.0002581966109573841
121    ],
122    "p_value": 0.0317,
123    "mde": 0.0004164460873821372,
124    "mde_rel_pct": 5.911335261472642,
125    "verdict": "idea better (significant)",
126    "system_worked": true
127  },
128  "custom_track": {
129    "name": "poisson_boundary",
130    "file": "fractional_dirichlet_track.py",
131    "domain": "pde"
132  },
133  "idea_sweep": [
134    {
135      "cfg": {
136        "lr": 0.001,
137        "lambda_g": 0.0
138      },
139      "mean": 0.013011600589379668
140    },
141    {
142      "cfg": {
143        "lr": 0.003,
144        "lambda_g": 0.001
145      },
146      "mean": 0.008521844982169569
147    },
148    {
149      "cfg": {
150        "lr": 0.01,
151        "lambda_g": 0.01
152      },
153      "mean": 0.0068476563319563866
154    }
155  ],
156  "protocol": {
157    "epochs": 35,
158    "batch": 128,
159    "a": 0.75,
160    "s": 0.1,
161    "delta": 1e-05,
162    "paired_seeds": [
163      0,
164      1,
165      2,
166      3,
167      4,
168      5,
169      6,
170      7
171    ]
172  },
173  "runtime_seconds": 58.30213451385498,
174  "mechanism_signature": {
175    "quantity": "median |u| at d=0.02 divided by median |u| at d=0.10 on trained models",
176    "predicted_idea_ratio": 0.2990697562442441,
177    "predicted_baseline_ratio": "not constrained (free output with boundary penalty)",
178    "observed_idea": 0.302967306501504,
179    "observed_baseline": 0.37791017847441105,
180    "confirmed": true
181  }
182}