OT Primitive Universal Flow / 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": 0.006,
 10      "weight_decay": 0.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.0015,
 16          "weight_decay": 0.0
 17        },
 18        "mean": 0.001414267288055271
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "weight_decay": 0.0
 24        },
 25        "mean": 0.0008843674877425656
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.006,
 30          "weight_decay": 0.0
 31        },
 32        "mean": 0.0005579828139161691
 33      }
 34    ],
 35    "full": {
 36      "mean": 0.0005214317206991836,
 37      "std": 7.105061588211062e-05,
 38      "per_seed": [
 39        0.0005374288302846253,
 40        0.00047665066085755825,
 41        0.0005716764717362821,
 42        0.0006461752927862108,
 43        0.0005541553255170584,
 44        0.0004454980371519923,
 45        0.00040709308814257383,
 46        0.0005327760591171682
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "best_cfg": {
 53      "lr": 0.006,
 54      "weight_decay": 0.0
 55    },
 56    "sweep": [
 57      {
 58        "cfg": {
 59          "lr": 0.0015,
 60          "weight_decay": 0.0
 61        },
 62        "mean": 0.0013487355754477903,
 63        "std": 0.0004249648894380296,
 64        "per_seed": [
 65          0.0011989021440967917,
 66          0.0010821575997397304,
 67          0.0014545865124091506,
 68          0.0019255068618804216,
 69          0.0007581147947348654,
 70          0.0009458610438741744,
 71          0.0013719621347263455,
 72          0.002052793512120843
 73        ],
 74        "n": 8
 75      },
 76      {
 77        "cfg": {
 78          "lr": 0.003,
 79          "weight_decay": 0.0
 80        },
 81        "mean": 0.0008453956179437228,
 82        "std": 0.00020597588608945252,
 83        "per_seed": [
 84          0.00080592290032655,
 85          0.0006897490820847452,
 86          0.0008771982393227518,
 87          0.0011676604626700282,
 88          0.0006385938613675535,
 89          0.0006526323850266635,
 90          0.0007422250346280634,
 91          0.0011891829781234264
 92        ],
 93        "n": 8
 94      },
 95      {
 96        "cfg": {
 97          "lr": 0.006,
 98          "weight_decay": 0.0
 99        },
100        "mean": 0.0005217095131229144,
101        "std": 7.135342401294591e-05,
102        "per_seed": [
103          0.0005381343653425574,
104          0.0004766805504914373,
105          0.000572634453419596,
106          0.0006471937522292137,
107          0.0005527037428691983,
108          0.0004458390176296234,
109          0.00040668679866939783,
110          0.0005338034243322909
111        ],
112        "n": 8
113      }
114    ],
115    "per_seed": [
116      0.0005381343653425574,
117      0.0004766805504914373,
118      0.000572634453419596,
119      0.0006471937522292137,
120      0.0005527037428691983,
121      0.0004458390176296234,
122      0.00040668679866939783,
123      0.0005338034243322909
124    ],
125    "mean": 0.0005217095131229144,
126    "std": 7.135342401294591e-05,
127    "n": 8
128  },
129  "comparison": {
130    "delta_mean": 2.777924237307161e-07,
131    "idea_wins": 2,
132    "n_pairs": 8,
133    "per_seed_diffs": [
134      7.055350579321384e-07,
135      2.9889633879065514e-08,
136      9.579816833138466e-07,
137      1.0184594430029392e-06,
138      -1.4515826478600502e-06,
139      3.4098047763109207e-07,
140      -4.062894731760025e-07,
141      1.0273652151226997e-06
142    ],
143    "p_value": 0.39955,
144    "mde": 7.262117301089291e-07,
145    "mde_rel_pct": 0.13927264132208098,
146    "verdict": "no measurable effect",
147    "system_worked": false,
148    "paired": {
149      "baseline_at_idea_cfg": {
150        "mean": 0.0005214317206991836,
151        "std": 7.105061588211062e-05,
152        "per_seed": [
153          0.0005374288302846253,
154          0.00047665066085755825,
155          0.0005716764717362821,
156          0.0006461752927862108,
157          0.0005541553255170584,
158          0.0004454980371519923,
159          0.00040709308814257383,
160          0.0005327760591171682
161        ],
162        "n": 8
163      },
164      "deltas_idea_minus_baseline": [
165        7.055350579321384e-07,
166        2.9889633879065514e-08,
167        9.579816833138466e-07,
168        1.0184594430029392e-06,
169        -1.4515826478600502e-06,
170        3.4098047763109207e-07,
171        -4.062894731760025e-07,
172        1.0273652151226997e-06
173      ],
174      "pvalue": 0.39955
175    }
176  },
177  "paired": {
178    "baseline_at_idea_cfg": {
179      "mean": 0.0005214317206991836,
180      "std": 7.105061588211062e-05,
181      "per_seed": [
182        0.0005374288302846253,
183        0.00047665066085755825,
184        0.0005716764717362821,
185        0.0006461752927862108,
186        0.0005541553255170584,
187        0.0004454980371519923,
188        0.00040709308814257383,
189        0.0005327760591171682
190      ],
191      "n": 8
192    },
193    "deltas_idea_minus_baseline": [
194      7.055350579321384e-07,
195      2.9889633879065514e-08,
196      9.579816833138466e-07,
197      1.0184594430029392e-06,
198      -1.4515826478600502e-06,
199      3.4098047763109207e-07,
200      -4.062894731760025e-07,
201      1.0273652151226997e-06
202    ],
203    "pvalue": 0.39955
204  },
205  "protocol": {
206    "epochs": 24,
207    "batch": 128,
208    "n_seeds": 8,
209    "loss": "MSE plus OT monotonicity and displacement penalties"
210  },
211  "mechanism_signature": {
212    "prediction": "OT monotonicity predicts positive terminal-angle Jacobian",
213    "idea": {
214      "predicted_monotone_fraction": 1.0,
215      "observed_monotone_fraction": 1.0,
216      "observed_mean_terminal_angle_jacobian": 0.24786332249641418,
217      "confirmed": true
218    },
219    "baseline": {
220      "predicted_monotone_fraction": 1.0,
221      "observed_monotone_fraction": 1.0,
222      "observed_mean_terminal_angle_jacobian": 0.2478588968515396,
223      "confirmed": true
224    },
225    "confirmed": true
226  }
227}