1{
  2  "bench_version": 1,
  3  "track": "vision",
  4  "model": "directional_cnn",
  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": 0.8287499845027924
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.8387499749660492
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.005
 27        },
 28        "mean": 0.8549999892711639
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.8506249710917473,
 33      "std": 0.038522525390718286,
 34      "per_seed": [
 35        0.7999999523162842,
 36        0.8849999904632568,
 37        0.824999988079071,
 38        0.7999999523162842,
 39        0.8399999737739563,
 40        0.85999995470047,
 41        0.9099999666213989,
 42        0.8849999904632568
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.8393749892711639,
 49    "std": 0.03591983491614504,
 50    "per_seed": [
 51      0.8149999976158142,
 52      0.8849999904632568,
 53      0.8050000071525574,
 54      0.7749999761581421,
 55      0.8399999737739563,
 56      0.8700000047683716,
 57      0.875,
 58      0.8499999642372131
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": -0.011249981820583344,
 64    "idea_wins": 4,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      0.01500004529953003,
 68      0.0,
 69      -0.019999980926513672,
 70      -0.02499997615814209,
 71      0.0,
 72      0.010000050067901611,
 73      -0.034999966621398926,
 74      -0.0350000262260437
 75    ],
 76    "p_value": 0.18755,
 77    "mde": 0.016685714835920434,
 78    "mde_rel_pct": 1.9615830010850617,
 79    "verdict": "no significant win",
 80    "system_worked": false
 81  },
 82  "math_check": {
 83    "geometric_ratio": 0.5,
 84    "expected_ratio": 0.5,
 85    "ratio_allclose": true,
 86    "K": 8,
 87    "M": 2,
 88    "path_split_counts": [
 89      3,
 90      3,
 91      3,
 92      3,
 93      3,
 94      3,
 95      3,
 96      3
 97    ],
 98    "max_splits": 3,
 99    "N": 3,
100    "bound_holds": true
101  },
102  "idea_sweep": [
103    {
104      "cfg": {
105        "lr": 0.001
106      },
107      "result": {
108        "mean": 0.8487499877810478,
109        "std": 0.04846068349664526,
110        "per_seed": [
111          0.7649999856948853,
112          0.8799999952316284,
113          0.8849999904632568,
114          0.7849999666213989,
115          0.8199999928474426,
116          0.8700000047683716,
117          0.9049999713897705,
118          0.8799999952316284
119        ],
120        "n": 8
121      }
122    },
123    {
124      "cfg": {
125        "lr": 0.003
126      },
127      "result": {
128        "mean": 0.8393749892711639,
129        "std": 0.03591983491614504,
130        "per_seed": [
131          0.8149999976158142,
132          0.8849999904632568,
133          0.8050000071525574,
134          0.7749999761581421,
135          0.8399999737739563,
136          0.8700000047683716,
137          0.875,
138          0.8499999642372131
139        ],
140        "n": 8
141      }
142    },
143    {
144      "cfg": {
145        "lr": 0.005
146      },
147      "result": {
148        "mean": 0.8443749845027924,
149        "std": 0.05405075357696703,
150        "per_seed": [
151          0.8999999761581421,
152          0.7400000095367432,
153          0.8299999833106995,
154          0.8299999833106995,
155          0.7899999618530273,
156          0.8799999952316284,
157          0.8999999761581421,
158          0.8849999904632568
159        ],
160        "n": 8
161      }
162    }
163  ],
164  "custom_track": null,
165  "structural_match": "spatial CIFAR images and convolutional directional fields",
166  "mechanism_signature": {
167    "prediction": "top-2 of K=8 uses 4x fewer directional samples",
168    "K": 8,
169    "beam": 2,
170    "predicted_reduction": 4.0,
171    "observed": [
172      {
173        "seed": 0,
174        "dense_metric": 0.7999999523162842,
175        "routed_metric": 0.7649999856948853,
176        "dense_samples_per_query": 12.0,
177        "routed_samples_per_query": 3.0,
178        "observed_reduction": 4.0,
179        "routing_entropy": 2.0645053386688232
180      },
181      {
182        "seed": 1,
183        "dense_metric": 0.8799999952316284,
184        "routed_metric": 0.8799999952316284,
185        "dense_samples_per_query": 12.0,
186        "routed_samples_per_query": 3.0,
187        "observed_reduction": 4.0,
188        "routing_entropy": 2.0756473541259766
189      }
190    ],
191    "confirmed": true
192  }
193}