Cross-Ratio Reversible Lattice Layer / 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": 0.003,
 10      "epochs": 25,
 11      "weight_decay": 0.0001
 12    },
 13    "sweep": [
 14      {
 15        "cfg": {
 16          "lr": 0.001,
 17          "epochs": 25,
 18          "weight_decay": 0.0
 19        },
 20        "mean": 0.006130807567387819
 21      },
 22      {
 23        "cfg": {
 24          "lr": 0.001,
 25          "epochs": 25,
 26          "weight_decay": 0.0001
 27        },
 28        "mean": 0.006120537058450282
 29      },
 30      {
 31        "cfg": {
 32          "lr": 0.003,
 33          "epochs": 25,
 34          "weight_decay": 0.0
 35        },
 36        "mean": 0.0046020153677091
 37      },
 38      {
 39        "cfg": {
 40          "lr": 0.003,
 41          "epochs": 25,
 42          "weight_decay": 0.0001
 43        },
 44        "mean": 0.0045645206701010466
 45      },
 46      {
 47        "cfg": {
 48          "lr": 0.01,
 49          "epochs": 25,
 50          "weight_decay": 0.0
 51        },
 52        "mean": 0.004882440727669746
 53      },
 54      {
 55        "cfg": {
 56          "lr": 0.01,
 57          "epochs": 25,
 58          "weight_decay": 0.0001
 59        },
 60        "mean": 0.004718217242043465
 61      }
 62    ],
 63    "full": {
 64      "mean": 0.004535878775641322,
 65      "std": 0.0004771483174727427,
 66      "per_seed": [
 67        0.004597439430654049,
 68        0.0038385987281799316,
 69        0.004727847874164581,
 70        0.005094196647405624,
 71        0.004155295435339212,
 72        0.00398264080286026,
 73        0.00463221687823534,
 74        0.005258794408291578
 75      ],
 76      "n": 8
 77    }
 78  },
 79  "idea": {
 80    "mean": 0.004785479104612023,
 81    "std": 0.00089246428498962,
 82    "per_seed": [
 83      0.004405152052640915,
 84      0.0037003301549702883,
 85      0.00573378661647439,
 86      0.005794585682451725,
 87      0.0041740648448467255,
 88      0.0035881984513252974,
 89      0.004965356085449457,
 90      0.005922358948737383
 91    ],
 92    "n": 8
 93  },
 94  "comparison": {
 95    "delta_mean": 0.0002496003289707005,
 96    "idea_wins": 3,
 97    "n_pairs": 8,
 98    "per_seed_diffs": [
 99      -0.000192287378013134,
100      -0.00013826857320964336,
101      0.0010059387423098087,
102      0.0007003890350461006,
103      1.8769409507513046e-05,
104      -0.00039444235153496265,
105      0.00033313920721411705,
106      0.0006635645404458046
107    ],
108    "p_value": 0.209,
109    "mde": 0.0004201483148548282,
110    "mde_rel_pct": 9.262776534309474,
111    "verdict": "no significant win",
112    "system_worked": false
113  },
114  "idea_sweep": [
115    {
116      "cfg": {
117        "lr": 0.003,
118        "epochs": 25,
119        "weight_decay": 0.0001
120      },
121      "result": {
122        "mean": 0.006392467068508267,
123        "std": 0.0020169420755724718,
124        "per_seed": [
125          0.005178917199373245,
126          0.01132385153323412,
127          0.006093441043049097,
128          0.0059410021640360355,
129          0.006832372397184372,
130          0.006475373636931181,
131          0.004938049241900444,
132          0.004356729332357645
133        ],
134        "n": 8
135      }
136    },
137    {
138      "cfg": {
139        "lr": 0.001,
140        "epochs": 25,
141        "weight_decay": 0.0
142      },
143      "result": {
144        "mean": 0.022261007456108928,
145        "std": 0.02048209560797128,
146        "per_seed": [
147          0.014691504649817944,
148          0.015150383114814758,
149          0.07583310455083847,
150          0.014610606245696545,
151          0.010224367491900921,
152          0.017106497660279274,
153          0.02021658793091774,
154          0.01025500800460577
155        ],
156        "n": 8
157      }
158    },
159    {
160      "cfg": {
161        "lr": 0.01,
162        "epochs": 25,
163        "weight_decay": 0.0001
164      },
165      "result": {
166        "mean": 0.004785479104612023,
167        "std": 0.00089246428498962,
168        "per_seed": [
169          0.004405152052640915,
170          0.0037003301549702883,
171          0.00573378661647439,
172          0.005794585682451725,
173          0.0041740648448467255,
174          0.0035881984513252974,
175          0.004965356085449457,
176          0.005922358948737383
177        ],
178        "n": 8
179      }
180    }
181  ],
182  "structural_match": "stability/control dynamics",
183  "mechanism_signature": {
184    "prediction": "trained model plaquette residual near zero",
185    "observed_mean_residual": 3.852884162824921e-07,
186    "observed_singular_fraction": 0.0,
187    "predicted_float32_scale": "~1e-6",
188    "confirmed": true
189  }
190}