Time-Shell Long-Horizon Decoder / bench_report.json

✓✓ Beats tuned baseline

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "multihorizon_sequence",
  4  "model": "custom_gru_decoder",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.006,
 10      "scale": 1.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "scale": 0.5
 17        },
 18        "mean": 0.36861784756183624
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "scale": 1.0
 24        },
 25        "mean": 0.3673807680606842
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.003,
 30          "scale": 0.5
 31        },
 32        "mean": 0.26081548631191254
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "scale": 1.0
 38        },
 39        "mean": 0.2626522071659565
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.006,
 44          "scale": 0.5
 45        },
 46        "mean": 0.2287113033235073
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.006,
 51          "scale": 1.0
 52        },
 53        "mean": 0.22365260869264603
 54      }
 55    ],
 56    "full": {
 57      "mean": 0.2238331437110901,
 58      "std": 0.021044464376645176,
 59      "per_seed": [
 60        0.2033109813928604,
 61        0.24181047081947327,
 62        0.23719552159309387,
 63        0.21229346096515656,
 64        0.2200879603624344,
 65        0.26243942975997925,
 66        0.220900297164917,
 67        0.19262702763080597
 68      ],
 69      "n": 8
 70    }
 71  },
 72  "idea": {
 73    "mean": 0.207510719075799,
 74    "std": 0.016642018572369865,
 75    "per_seed": [
 76      0.20119042694568634,
 77      0.23540125787258148,
 78      0.23484113812446594,
 79      0.2042408287525177,
 80      0.20302963256835938,
 81      0.1936691254377365,
 82      0.19913142919540405,
 83      0.1885819137096405
 84    ],
 85    "n": 8
 86  },
 87  "comparison": {
 88    "delta_mean": -0.0163224246352911,
 89    "idea_wins": 8,
 90    "n_pairs": 8,
 91    "per_seed_diffs": [
 92      -0.0021205544471740723,
 93      -0.006409212946891785,
 94      -0.0023543834686279297,
 95      -0.008052632212638855,
 96      -0.017058327794075012,
 97      -0.06877030432224274,
 98      -0.02176886796951294,
 99      -0.004045113921165466
100    ],
101    "p_value": 0.0081,
102    "mde": 0.018684083393322008,
103    "mde_rel_pct": 8.347326532409454,
104    "verdict": "idea better (significant)",
105    "system_worked": true
106  },
107  "mechanism_signature": {
108    "signature": {
109      "prediction": "reverse cumulative shell makes early outputs more sensitive to later coefficients",
110      "predicted": {
111        "early_over_late_sensitivity": ">1"
112      },
113      "observed": {
114        "early_abs_change": 0.0,
115        "late_abs_change": 0.016314398497343063,
116        "early_over_late_sensitivity": 0.0,
117        "trained_test_mse": 0.20119042694568634
118      },
119      "confirmed": false,
120      "forward_seconds": 1.733663829996658
121    },
122    "idea_sweep": [
123      {
124        "cfg": {
125          "lr": 0.006,
126          "scale": 1.0
127        },
128        "result": {
129          "mean": 0.207510719075799,
130          "std": 0.016642018572369865,
131          "per_seed": [
132            0.20119042694568634,
133            0.23540125787258148,
134            0.23484113812446594,
135            0.2042408287525177,
136            0.20302963256835938,
137            0.1936691254377365,
138            0.19913142919540405,
139            0.1885819137096405
140          ],
141          "n": 8
142        }
143      },
144      {
145        "cfg": {
146          "lr": 0.001,
147          "scale": 1.0
148        },
149        "result": {
150          "mean": 0.339939147233963,
151          "std": 0.038412014687187156,
152          "per_seed": [
153            0.35680338740348816,
154            0.38626959919929504,
155            0.4020107686519623,
156            0.31106022000312805,
157            0.35761505365371704,
158            0.29985377192497253,
159            0.2990495562553406,
160            0.3068508207798004
161          ],
162          "n": 8
163        }
164      },
165      {
166        "cfg": {
167          "lr": 0.006,
168          "scale": 0.5
169        },
170        "result": {
171          "mean": 0.21021396107971668,
172          "std": 0.022721494928976002,
173          "per_seed": [
174            0.21364261209964752,
175            0.23506788909435272,
176            0.23733554780483246,
177            0.20955374836921692,
178            0.2235165685415268,
179            0.21272434294223785,
180            0.1817266047000885,
181            0.1681443750858307
182          ],
183          "n": 8
184        }
185      }
186    ],
187    "complexity_prediction": {
188      "shell": "O(Kd)",
189      "dense": "O(K^2d)"
190    },
191    "custom_track": {
192      "name": "multihorizon_sequence",
193      "file": "custom_track.py",
194      "domain": "sequence"
195    }
196  }
197}