Compactified Burst Controller / bench_report.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "dynamics",
  4  "model": "rnn_small",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.009,
 10      "epochs": 12,
 11      "weight_decay": 0.0001
 12    },
 13    "sweep": [
 14      {
 15        "cfg": {
 16          "lr": 0.001,
 17          "epochs": 12,
 18          "weight_decay": 0.0
 19        },
 20        "mean": 0.004103701037820429
 21      },
 22      {
 23        "cfg": {
 24          "lr": 0.001,
 25          "epochs": 12,
 26          "weight_decay": 0.0001
 27        },
 28        "mean": 0.0041049906285479665
 29      },
 30      {
 31        "cfg": {
 32          "lr": 0.003,
 33          "epochs": 12,
 34          "weight_decay": 0.0
 35        },
 36        "mean": 0.0033861721749417484
 37      },
 38      {
 39        "cfg": {
 40          "lr": 0.003,
 41          "epochs": 12,
 42          "weight_decay": 0.0001
 43        },
 44        "mean": 0.0033592699037399143
 45      },
 46      {
 47        "cfg": {
 48          "lr": 0.009,
 49          "epochs": 12,
 50          "weight_decay": 0.0
 51        },
 52        "mean": 0.0012523236509878188
 53      },
 54      {
 55        "cfg": {
 56          "lr": 0.009,
 57          "epochs": 12,
 58          "weight_decay": 0.0001
 59        },
 60        "mean": 0.0012439452111721039
 61      }
 62    ],
 63    "full": {
 64      "mean": 0.0015301889216061682,
 65      "std": 0.000520398293562222,
 66      "per_seed": [
 67        0.0006319124950096011,
 68        0.0014439183287322521,
 69        0.0010583436815068126,
 70        0.0018416063394397497,
 71        0.0024377191439270973,
 72        0.0017625956097617745,
 73        0.0018110211240127683,
 74        0.0012543946504592896
 75      ],
 76      "n": 8
 77    }
 78  },
 79  "idea": {
 80    "mean": 0.004236663109622896,
 81    "std": 0.0018448852809520508,
 82    "per_seed": [
 83      0.0038791601546108723,
 84      0.006286025978624821,
 85      0.003358118701726198,
 86      0.007666262332350016,
 87      0.0023571187630295753,
 88      0.0020544668659567833,
 89      0.003132161218672991,
 90      0.0051599908620119095
 91    ],
 92    "n": 8
 93  },
 94  "comparison": {
 95    "delta_mean": 0.0027064741880167276,
 96    "idea_wins": 1,
 97    "n_pairs": 8,
 98    "per_seed_diffs": [
 99      0.003247247659601271,
100      0.004842107649892569,
101      0.0022997750202193856,
102      0.005824655992910266,
103      -8.060038089752197e-05,
104      0.00029187125619500875,
105      0.0013211400946602225,
106      0.00390559621155262
107    ],
108    "p_value": 0.01475,
109    "mde": 0.0017792868429345541,
110    "mde_rel_pct": 116.27889980192245,
111    "verdict": "idea worse (significant)",
112    "system_worked": false
113  },
114  "mechanism_signature": {
115    "mechanism_signature": {
116      "predicted_effect": "radial damping lowers burst norm when positive radial growth and near-equilibrium gate coincide",
117      "observed_mean_baseline_max_norm": 2.6786648333072662,
118      "observed_mean_idea_max_norm": 2.058234691619873,
119      "mean_norm_reduction": 0.6204301416873932,
120      "total_interventions": 1919,
121      "confirmed": true
122    }
123  },
124  "idea_sweep": [
125    {
126      "cfg": {
127        "lr": 0.009,
128        "epochs": 12,
129        "weight_decay": 0.0001
130      },
131      "result": {
132        "mean": 0.004236663109622896,
133        "std": 0.0018448852809520508,
134        "per_seed": [
135          0.0038791601546108723,
136          0.006286025978624821,
137          0.003358118701726198,
138          0.007666262332350016,
139          0.0023571187630295753,
140          0.0020544668659567833,
141          0.003132161218672991,
142          0.0051599908620119095
143        ],
144        "n": 8
145      }
146    },
147    {
148      "cfg": {
149        "lr": 0.001,
150        "epochs": 12,
151        "weight_decay": 0.0001
152      },
153      "result": {
154        "mean": 0.05461519444361329,
155        "std": 0.009953145481423375,
156        "per_seed": [
157          0.047671690583229065,
158          0.055415134876966476,
159          0.04205615073442459,
160          0.06574815511703491,
161          0.044831324368715286,
162          0.04755406454205513,
163          0.0637914165854454,
164          0.06985361874103546
165        ],
166        "n": 8
167      }
168    },
169    {
170      "cfg": {
171        "lr": 0.009,
172        "epochs": 12,
173        "weight_decay": 0.0001
174      },
175      "result": {
176        "mean": 0.004236663109622896,
177        "std": 0.0018448852809520508,
178        "per_seed": [
179          0.0038791601546108723,
180          0.006286025978624821,
181          0.003358118701726198,
182          0.007666262332350016,
183          0.0023571187630295753,
184          0.0020544668659567833,
185          0.003132161218672991,
186          0.0051599908620119095
187        ],
188        "n": 8
189      }
190    }
191  ],
192  "device": "cuda"
193}