Chern-Gap Monitor for Finite-Horizon Collapse / bench_output.json
Mechanism confirmed, baseline not beaten
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 },
11 "sweep": [
12 {
13 "cfg": {
14 "lr": 0.001
15 },
16 "mean": 1.596962183713913
17 },
18 {
19 "cfg": {
20 "lr": 0.003
21 },
22 "mean": 1.552410066127777
23 },
24 {
25 "cfg": {
26 "lr": 0.01
27 },
28 "mean": 1.5703844726085663
29 }
30 ],
31 "full": {
32 "mean": 1.5021538734436035,
33 "std": 0.09324438560832148,
34 "per_seed": [
35 1.4778438806533813,
36 1.4308412075042725,
37 1.5888422727584839,
38 1.7121129035949707,
39 1.4590684175491333,
40 1.4697362184524536,
41 1.4667506217956543,
42 1.4120354652404785
43 ],
44 "n": 8
45 }
46 },
47 "idea": {
48 "mean": 1.5022293776273727,
49 "std": 0.09287950268012642,
50 "per_seed": [
51 1.4786245822906494,
52 1.4313784837722778,
53 1.5867565870285034,
54 1.7121129035949707,
55 1.459558367729187,
56 1.4695805311203003,
57 1.4680730104446411,
58 1.4117505550384521
59 ],
60 "n": 8
61 },
62 "comparison": {
63 "delta_mean": 7.550418376922607e-05,
64 "idea_wins": 3,
65 "n_pairs": 8,
66 "per_seed_diffs": [
67 0.0007807016372680664,
68 0.0005372762680053711,
69 -0.0020856857299804688,
70 0.0,
71 0.0004899501800537109,
72 -0.0001556873321533203,
73 0.0013223886489868164,
74 -0.0002849102020263672
75 ],
76 "p_value": 0.82525,
77 "mde": 0.0008536970299377886,
78 "mde_rel_pct": 0.05683153004696756,
79 "verdict": "no measurable effect",
80 "system_worked": false
81 },
82 "idea_sweep": [
83 {
84 "cfg": {
85 "lr": 0.003
86 },
87 "result": {
88 "mean": 1.5022293776273727,
89 "std": 0.09287950268012642,
90 "per_seed": [
91 1.4786245822906494,
92 1.4313784837722778,
93 1.5867565870285034,
94 1.7121129035949707,
95 1.459558367729187,
96 1.4695805311203003,
97 1.4680730104446411,
98 1.4117505550384521
99 ],
100 "n": 8
101 }
102 },
103 {
104 "cfg": {
105 "lr": 0.001
106 },
107 "result": {
108 "mean": 1.5371521562337875,
109 "std": 0.10299458835449656,
110 "per_seed": [
111 1.533307433128357,
112 1.483575701713562,
113 1.607698917388916,
114 1.7639100551605225,
115 1.5047532320022583,
116 1.5342274904251099,
117 1.4773060083389282,
118 1.3924384117126465
119 ],
120 "n": 8
121 }
122 },
123 {
124 "cfg": {
125 "lr": 0.01
126 },
127 "result": {
128 "mean": 1.5104802548885345,
129 "std": 0.10079049945434744,
130 "per_seed": [
131 1.4905154705047607,
132 1.4462878704071045,
133 1.600508213043213,
134 1.7361385822296143,
135 1.4536124467849731,
136 1.5005030632019043,
137 1.448508620262146,
138 1.4077677726745605
139 ],
140 "n": 8
141 }
142 }
143 ],
144 "track_justification": "dynamics matches stability/control structure of the finite-horizon collapse monitor",
145 "mechanism_signature": {
146 "probe": "trained RNN predictions under three fixed phase perturbations",
147 "per_seed": [
148 {
149 "seed": 0,
150 "baseline_metric": 1.4778438806533813,
151 "idea_metric": 1.4786245822906494,
152 "baseline_gap": 0.0034411137457937002,
153 "idea_gap": 0.005794507451355457,
154 "baseline_chern": 1.7380369271595016e-08,
155 "idea_chern": -4.784160125544356e-10
156 },
157 {
158 "seed": 1,
159 "baseline_metric": 1.4308412075042725,
160 "idea_metric": 1.4313784837722778,
161 "baseline_gap": 0.08322877436876297,
162 "idea_gap": 0.08180162310600281,
163 "baseline_chern": -5.830075504012453e-10,
164 "idea_chern": 2.12119036466648e-09
165 },
166 {
167 "seed": 2,
168 "baseline_metric": 1.5888422727584839,
169 "idea_metric": 1.5867565870285034,
170 "baseline_gap": 0.003892321139574051,
171 "idea_gap": 0.004510408267378807,
172 "baseline_chern": -1.4004964258741766e-09,
173 "idea_chern": -8.674661327411772e-09
174 },
175 {
176 "seed": 3,
177 "baseline_metric": 1.7121129035949707,
178 "idea_metric": 1.7121129035949707,
179 "baseline_gap": 0.08939465880393982,
180 "idea_gap": 0.08939465880393982,
181 "baseline_chern": -6.488612319724056e-11,
182 "idea_chern": -6.488612319724056e-11
183 },
184 {
185 "seed": 4,
186 "baseline_metric": 1.4590684175491333,
187 "idea_metric": 1.459558367729187,
188 "baseline_gap": 0.03839415684342384,
189 "idea_gap": 0.049591924995183945,
190 "baseline_chern": -2.129452757791829e-09,
191 "idea_chern": 5.6372739935792205e-09
192 },
193 {
194 "seed": 5,
195 "baseline_metric": 1.4697362184524536,
196 "idea_metric": 1.4695805311203003,
197 "baseline_gap": 0.10124503821134567,
198 "idea_gap": 0.10104020684957504,
199 "baseline_chern": -4.887011095528017e-09,
200 "idea_chern": -2.0993884286090966e-09
201 },
202 {
203 "seed": 6,
204 "baseline_metric": 1.4667506217956543,
205 "idea_metric": 1.4680730104446411,
206 "baseline_gap": 0.079883873462677,
207 "idea_gap": 0.07974080741405487,
208 "baseline_chern": -9.885893837729884e-09,
209 "idea_chern": 1.0790899473853646e-09
210 },
211 {
212 "seed": 7,
213 "baseline_metric": 1.4120354652404785,
214 "idea_metric": 1.4117505550384521,
215 "baseline_gap": 0.2863689661026001,
216 "idea_gap": 0.28592658042907715,
217 "baseline_chern": 2.0020123666392965e-09,
218 "idea_chern": 6.683505491112086e-11
219 }
220 ],
221 "training_checkpoints_seed0": [
222 {
223 "epoch": 1,
224 "gap": 0.0251762717962265,
225 "chern": 2.9378251026121728e-09,
226 "rounded_chern": 0
227 },
228 {
229 "epoch": 3,
230 "gap": 0.058473870158195496,
231 "chern": 2.4474322762181137e-10,
232 "rounded_chern": 0
233 },
234 {
235 "epoch": 5,
236 "gap": 0.054124489426612854,
237 "chern": -3.3403893006553517e-09,
238 "rounded_chern": 0
239 },
240 {
241 "epoch": 7,
242 "gap": 0.01788416877388954,
243 "chern": -3.5472674164821973e-09,
244 "rounded_chern": 0
245 },
246 {
247 "epoch": 9,
248 "gap": 0.015780657529830933,
249 "chern": -4.577209723398702e-09,
250 "rounded_chern": 0
251 },
252 {
253 "epoch": 10,
254 "gap": 0.005794507451355457,
255 "chern": -4.784160125544356e-10,
256 "rounded_chern": 0
257 }
258 ],
259 "predicted": "Chern stays constant while gap is nonzero; sector changes require gap closing",
260 "observed_transition_count": 0,
261 "min_checkpoint_gap": 0.005794507451355457,
262 "confirmed": true
263 }
264}