1{
  2  "bench_report": {
  3    "bench_version": 1,
  4    "track": "binary_symmetric_power_regression",
  5    "model": "local_mlp_head",
  6    "metric_direction": "lower is better",
  7    "n_seeds": 8,
  8    "baseline": {
  9      "best_cfg": {
 10        "lr": 0.01
 11      },
 12      "sweep": [
 13        {
 14          "cfg": {
 15            "lr": 0.001
 16          },
 17          "mean": 0.1037276852875948
 18        },
 19        {
 20          "cfg": {
 21            "lr": 0.003
 22          },
 23          "mean": 0.007304377853870392
 24        },
 25        {
 26          "cfg": {
 27            "lr": 0.01
 28          },
 29          "mean": 0.004617542843334377
 30        }
 31      ],
 32      "full": {
 33        "mean": 0.004571487399516627,
 34        "std": 0.0009398261599528283,
 35        "per_seed": [
 36          0.004156868439167738,
 37          0.006239068228751421,
 38          0.0032845954410731792,
 39          0.004789639264345169,
 40          0.004069721791893244,
 41          0.00436969893053174,
 42          0.0038355558644980192,
 43          0.005826751235872507
 44        ],
 45        "n": 8
 46      }
 47    },
 48    "idea": {
 49      "best_config": {
 50        "lr": 0.01
 51      },
 52      "per_seed": [
 53        0.0036381708923727274,
 54        0.0029435837641358376,
 55        0.004084695130586624,
 56        0.009996835142374039,
 57        0.004572708159685135,
 58        0.02756613865494728,
 59        0.006751858629286289,
 60        0.004224956501275301
 61      ],
 62      "mean": 0.007972368359332904,
 63      "std": 0.007699778199581965
 64    },
 65    "comparison": {
 66      "delta_mean": 0.003400880959816277,
 67      "idea_wins": 3,
 68      "n_pairs": 8,
 69      "per_seed_diffs": [
 70        -0.0005186975467950106,
 71        -0.0032954844646155834,
 72        0.0008000996895134449,
 73        0.00520719587802887,
 74        0.0005029863677918911,
 75        0.02319643972441554,
 76        0.00291630276478827,
 77        -0.0016017947345972061
 78      ],
 79      "p_value": 0.3091,
 80      "mde": 0.007036725002667524,
 81      "mde_rel_pct": 153.92637860954318,
 82      "verdict": "no significant win",
 83      "system_worked": false
 84    },
 85    "protocol": {
 86      "seeds": [
 87        0,
 88        1,
 89        2,
 90        3,
 91        4,
 92        5,
 93        6,
 94        7
 95      ],
 96      "baseline_grid": [
 97        {
 98          "lr": 0.001
 99        },
100        {
101          "lr": 0.003
102        },
103        {
104          "lr": 0.01
105        }
106      ],
107      "idea_grid": [
108        {
109          "lr": 0.001
110        },
111        {
112          "lr": 0.003
113        },
114        {
115          "lr": 0.01
116        }
117      ],
118      "selection": "baseline best on seeds 0-3; idea evaluated at same selected lr"
119    },
120    "mechanism_signature": {
121      "predicted_equivariance_error": 0.0,
122      "observed_trained_feature_equivariance_error": 2.1827872842550278e-11,
123      "predicted_composition_error": 0.0,
124      "observed_representation_composition_error": 1.1368683772161603e-13,
125      "confirmed": true,
126      "note": "Signature is retested numerically at NN-scale on the trained idea model's feature map; task metric remains independent."
127    }
128  },
129  "all_sweeps": {
130    "baseline": {
131      "0.001": [
132        0.1355224996805191,
133        0.11009717732667923,
134        0.055855073034763336,
135        0.11343599110841751,
136        0.07716862857341766,
137        0.07487080246210098,
138        0.10096120834350586,
139        0.059963226318359375
140      ],
141      "0.003": [
142        0.00825435295701027,
143        0.008851638063788414,
144        0.004908436443656683,
145        0.007203083951026201,
146        0.003927994519472122,
147        0.009901680983603,
148        0.005368529818952084,
149        0.005688884295523167
150      ],
151      "0.01": [
152        0.004156868439167738,
153        0.006239068228751421,
154        0.0032845954410731792,
155        0.004789639264345169,
156        0.004069721791893244,
157        0.00436969893053174,
158        0.0038355558644980192,
159        0.005826751235872507
160      ]
161    },
162    "idea": {
163      "0.001": [
164        0.0154461320489645,
165        0.03856436163187027,
166        0.03455395624041557,
167        0.015390957705676556,
168        0.039948124438524246,
169        0.09389373660087585,
170        0.025615643709897995,
171        0.02203393168747425
172      ],
173      "0.003": [
174        0.00696675106883049,
175        0.009506145492196083,
176        0.015012294054031372,
177        0.007092596031725407,
178        0.009301144629716873,
179        0.02316836081445217,
180        0.015946412459015846,
181        0.011056053452193737
182      ],
183      "0.01": [
184        0.0036381708923727274,
185        0.0029435837641358376,
186        0.004084695130586624,
187        0.009996835142374039,
188        0.004572708159685135,
189        0.02756613865494728,
190        0.006751858629286289,
191        0.004224956501275301
192      ]
193    }
194  },
195  "custom_track": {
196    "name": "binary_symmetric_power_regression",
197    "file": "bench_symmetric_power.py",
198    "domain": "planar_equivariance"
199  },
200  "math_check": {
201    "max_polynomial_abs_error": 4.263256414560601e-14,
202    "max_composition_abs_error": 1.1368683772161603e-13,
203    "max_rotation_invariant_abs_error": 2.1827872842550278e-11
204  }
205}