Certified Active-Tail Ising Layer / artifacts/bench_report.json
Failed on benchmark
1{
2 "bench_version": 1,
3 "track": "tabular",
4 "model": "mlp_tiny",
5 "metric_direction": "lower is better",
6 "n_seeds": 8,
7 "baseline": {
8 "sweep": {
9 "grid": [
10 {
11 "config": {
12 "lr": 0.001,
13 "steps": 3
14 },
15 "per_seed": [
16 178.24566650390625,
17 163.57577514648438,
18 174.44857788085938,
19 176.0863494873047
20 ],
21 "mean": 173.08909225463867
22 },
23 {
24 "config": {
25 "lr": 0.003,
26 "steps": 4
27 },
28 "per_seed": [
29 102.84964752197266,
30 99.15052795410156,
31 99.76235961914062,
32 110.30470275878906
33 ],
34 "mean": 103.01680946350098
35 },
36 {
37 "config": {
38 "lr": 0.01,
39 "steps": 5
40 },
41 "per_seed": [
42 28.39447021484375,
43 22.11431121826172,
44 28.96165657043457,
45 24.90264320373535
46 ],
47 "mean": 26.093270301818848
48 }
49 ],
50 "best": {
51 "config": {
52 "lr": 0.01,
53 "steps": 5
54 },
55 "per_seed": [
56 28.39447021484375,
57 22.11431121826172,
58 28.96165657043457,
59 24.90264320373535
60 ],
61 "mean": 26.093270301818848
62 }
63 },
64 "full": {
65 "config": {
66 "lr": 0.01,
67 "steps": 5
68 },
69 "per_seed": [
70 28.39447021484375,
71 22.11431121826172,
72 28.96165657043457,
73 24.90264320373535,
74 27.14609718322754,
75 25.709123611450195,
76 26.396347045898438,
77 29.079734802246094
78 ],
79 "mean": 26.588047981262207,
80 "signature_samples": [
81 {
82 "active_fraction": 1.0,
83 "frozen_fraction": 0.0,
84 "certification_violations": 0
85 },
86 {
87 "active_fraction": 1.0,
88 "frozen_fraction": 0.0,
89 "certification_violations": 0
90 },
91 {
92 "active_fraction": 1.0,
93 "frozen_fraction": 0.0,
94 "certification_violations": 0
95 },
96 {
97 "active_fraction": 1.0,
98 "frozen_fraction": 0.0,
99 "certification_violations": 0
100 },
101 {
102 "active_fraction": 1.0,
103 "frozen_fraction": 0.0,
104 "certification_violations": 0
105 },
106 {
107 "active_fraction": 1.0,
108 "frozen_fraction": 0.0,
109 "certification_violations": 0
110 },
111 {
112 "active_fraction": 1.0,
113 "frozen_fraction": 0.0,
114 "certification_violations": 0
115 },
116 {
117 "active_fraction": 1.0,
118 "frozen_fraction": 0.0,
119 "certification_violations": 0
120 }
121 ]
122 }
123 },
124 "idea": {
125 "config": {
126 "lr": 0.01,
127 "steps": 5
128 },
129 "per_seed": [
130 28.394472122192383,
131 22.11431121826172,
132 28.96165657043457,
133 24.90264129638672,
134 27.116384506225586,
135 25.70911979675293,
136 26.396347045898438,
137 29.079734802246094
138 ],
139 "mean": 26.584333419799805,
140 "signature_samples": [
141 {
142 "active_fraction": 1.0,
143 "frozen_fraction": 0.0,
144 "certification_violations": 0
145 },
146 {
147 "active_fraction": 1.0,
148 "frozen_fraction": 0.0,
149 "certification_violations": 0
150 },
151 {
152 "active_fraction": 1.0,
153 "frozen_fraction": 0.0,
154 "certification_violations": 0
155 },
156 {
157 "active_fraction": 1.0,
158 "frozen_fraction": 0.0,
159 "certification_violations": 0
160 },
161 {
162 "active_fraction": 0.96875,
163 "frozen_fraction": 0.03125,
164 "certification_violations": 0
165 },
166 {
167 "active_fraction": 1.0,
168 "frozen_fraction": 0.0,
169 "certification_violations": 0
170 },
171 {
172 "active_fraction": 1.0,
173 "frozen_fraction": 0.0,
174 "certification_violations": 0
175 },
176 {
177 "active_fraction": 1.0,
178 "frozen_fraction": 0.0,
179 "certification_violations": 0
180 }
181 ]
182 },
183 "comparison": {
184 "delta_mean": -0.0037145614624023438,
185 "idea_wins": 3,
186 "n_pairs": 8,
187 "per_seed_diffs": [
188 1.9073486328125e-06,
189 0.0,
190 0.0,
191 -1.9073486328125e-06,
192 -0.029712677001953125,
193 -3.814697265625e-06,
194 0.0,
195 0.0
196 ],
197 "p_value": 0.37185,
198 "mde": 0.008783649148145774,
199 "mde_rel_pct": 0.033036081303659474,
200 "verdict": "no measurable effect",
201 "system_worked": false
202 },
203 "math_sanity": {
204 "conditional_identity_max_abs_error": 1.7763568394002505e-15,
205 "exhaustive_certification_violations": 0,
206 "certification_bound_holds": true
207 },
208 "idea_sweep": {
209 "grid": [
210 {
211 "config": {
212 "lr": 0.001,
213 "steps": 3
214 },
215 "per_seed": [
216 178.24566650390625,
217 163.57577514648438,
218 174.44857788085938,
219 176.0863494873047
220 ],
221 "mean": 173.08909225463867
222 },
223 {
224 "config": {
225 "lr": 0.003,
226 "steps": 4
227 },
228 "per_seed": [
229 102.84964752197266,
230 99.15052795410156,
231 99.76235961914062,
232 110.30470275878906
233 ],
234 "mean": 103.01680946350098
235 },
236 {
237 "config": {
238 "lr": 0.01,
239 "steps": 5
240 },
241 "per_seed": [
242 28.394472122192383,
243 22.11431121826172,
244 28.96165657043457,
245 24.90264129638672
246 ],
247 "mean": 26.093270301818848
248 }
249 ],
250 "best": {
251 "config": {
252 "lr": 0.01,
253 "steps": 5
254 },
255 "per_seed": [
256 28.394472122192383,
257 22.11431121826172,
258 28.96165657043457,
259 24.90264129638672
260 ],
261 "mean": 26.093270301818848
262 }
263 },
264 "protocol_note": "Baseline and idea share IsingMLP architecture; only update rule differs.",
265 "mechanism_signature": {
266 "prediction": "certified active-tail reduces active interaction work after polarization",
267 "predicted_active_fraction_upper_bound": 1.0,
268 "observed_idea_active_fraction": 0.99609375,
269 "observed_baseline_active_fraction": 1.0,
270 "reduction_ratio": 0.99609375,
271 "confirmed": false
272 }
273}