1{
  2  "bench_version": 1,
  3  "track": "dynamics",
  4  "model": "factor_rnn",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.006,
 10      "epochs": 18,
 11      "beta": 0.0
 12    },
 13    "sweep": [
 14      {
 15        "cfg": {
 16          "lr": 0.0015,
 17          "epochs": 18,
 18          "beta": 0.0
 19        },
 20        "mean": 0.11639470420777798
 21      },
 22      {
 23        "cfg": {
 24          "lr": 0.003,
 25          "epochs": 18,
 26          "beta": 0.0
 27        },
 28        "mean": 0.058107024524360895
 29      },
 30      {
 31        "cfg": {
 32          "lr": 0.006,
 33          "epochs": 18,
 34          "beta": 0.0
 35        },
 36        "mean": 0.026168195763602853
 37      }
 38    ],
 39    "full": {
 40      "mean": 0.02109241997823119,
 41      "std": 0.014265532680015599,
 42      "per_seed": [
 43        0.05811716988682747,
 44        0.019254598766565323,
 45        0.013698350638151169,
 46        0.01360266376286745,
 47        0.014953133650124073,
 48        0.012289082631468773,
 49        0.021011028438806534,
 50        0.015813332051038742
 51      ],
 52      "n": 8
 53    },
 54    "full_all_union": [
 55      {
 56        "cfg": {
 57          "lr": 0.0015,
 58          "epochs": 18,
 59          "beta": 0.0
 60        },
 61        "mean": 0.08842666260898113,
 62        "std": 0.06345898927987136,
 63        "per_seed": [
 64          0.2491457462310791,
 65          0.04528465121984482,
 66          0.08107621967792511,
 67          0.09007219970226288,
 68          0.05925949290394783,
 69          0.04771618917584419,
 70          0.09064487367868423,
 71          0.044213928282260895
 72        ],
 73        "n": 8
 74      },
 75      {
 76        "cfg": {
 77          "lr": 0.003,
 78          "epochs": 18,
 79          "beta": 0.0
 80        },
 81        "mean": 0.04415802680887282,
 82        "std": 0.03152132845115148,
 83        "per_seed": [
 84          0.1255890429019928,
 85          0.02912258543074131,
 86          0.03296443447470665,
 87          0.04475203529000282,
 88          0.029448043555021286,
 89          0.025108061730861664,
 90          0.041623830795288086,
 91          0.024656180292367935
 92        ],
 93        "n": 8
 94      },
 95      {
 96        "cfg": {
 97          "lr": 0.006,
 98          "epochs": 18,
 99          "beta": 0.0
100        },
101        "mean": 0.02109241997823119,
102        "std": 0.014265532680015599,
103        "per_seed": [
104          0.05811716988682747,
105          0.019254598766565323,
106          0.013698350638151169,
107          0.01360266376286745,
108          0.014953133650124073,
109          0.012289082631468773,
110          0.021011028438806534,
111          0.015813332051038742
112        ],
113        "n": 8
114      }
115    ]
116  },
117  "idea": {
118    "mean": 0.022512523224577308,
119    "std": 0.012331270086282655,
120    "per_seed": [
121      0.04500043764710426,
122      0.017105672508478165,
123      0.03219568356871605,
124      0.034598458558321,
125      0.01256189402192831,
126      0.008819940499961376,
127      0.019357675686478615,
128      0.010460423305630684
129    ],
130    "n": 8
131  },
132  "comparison": {
133    "delta_mean": 0.001420103246346116,
134    "idea_wins": 6,
135    "n_pairs": 8,
136    "per_seed_diffs": [
137      -0.013116732239723206,
138      -0.002148926258087158,
139      0.01849733293056488,
140      0.02099579479545355,
141      -0.0023912396281957626,
142      -0.0034691421315073967,
143      -0.001653352752327919,
144      -0.005352908745408058
145    ],
146    "p_value": 0.78325,
147    "mde": 0.009957143075416026,
148    "mde_rel_pct": 47.20721038976311,
149    "verdict": "no significant win",
150    "system_worked": false
151  },
152  "mechanism_signature": {
153    "track_structure": "controlled pendulum multi-step dynamics",
154    "metric": 0.04500043764710426,
155    "max_component_violation": -0.35021981596946716,
156    "max_observed_ratio": 0.7172038555145264,
157    "predicted_bound": "nonpositive violation",
158    "observed_envelope_row_sum": 2.4505488872528076,
159    "observed_sum_abs_A": 45.71932601928711,
160    "observed_sum_P": 65.28011322021484,
161    "confirmed": true,
162    "config": {
163      "lr": 0.006,
164      "epochs": 18,
165      "beta": 0.02,
166      "target": 0.92
167    },
168    "wall_seconds": 27.75654865699471
169  },
170  "idea_grid": [
171    {
172      "cfg": {
173        "lr": 0.0015,
174        "epochs": 18,
175        "beta": 0.005,
176        "target": 0.92
177      },
178      "mean": 0.09474506694823503,
179      "std": 0.10094635313979229,
180      "per_seed": [
181        0.35983920097351074,
182        0.059099163860082626,
183        0.058137185871601105,
184        0.07874950021505356,
185        0.054427292197942734,
186        0.044902503490448,
187        0.0674169585108757,
188        0.035388730466365814
189      ],
190      "n": 8
191    },
192    {
193      "cfg": {
194        "lr": 0.003,
195        "epochs": 18,
196        "beta": 0.01,
197        "target": 0.92
198      },
199      "mean": 0.036793723003938794,
200      "std": 0.02819267034786451,
201      "per_seed": [
202        0.1090393140912056,
203        0.0328064002096653,
204        0.026808833703398705,
205        0.040870074182748795,
206        0.023899797350168228,
207        0.01856924220919609,
208        0.024410061538219452,
209        0.017946060746908188
210      ],
211      "n": 8
212    },
213    {
214      "cfg": {
215        "lr": 0.006,
216        "epochs": 18,
217        "beta": 0.02,
218        "target": 0.92
219      },
220      "mean": 0.022512523224577308,
221      "std": 0.012331270086282655,
222      "per_seed": [
223        0.04500043764710426,
224        0.017105672508478165,
225        0.03219568356871605,
226        0.034598458558321,
227        0.01256189402192831,
228        0.008819940499961376,
229        0.019357675686478615,
230        0.010460423305630684
231      ],
232      "n": 8
233    }
234  ],
235  "device": "cuda",
236  "note": "Baseline and idea are end-to-end trained systems with identical FactorRNN architecture; only envelope loss differs. All shared learning rates were evaluated on the same eight paired seeds."
237}