Reachable-Set Risk Head for Early-Warning Rollouts / results.json
Failed on benchmark
1{
2 "seed": 1449,
3 "mechanism": {
4 "margin_sweep": {
5 "margins": [
6 -3.0,
7 -2.0,
8 -1.0,
9 0.0,
10 1.0,
11 2.0,
12 3.0
13 ],
14 "q": [
15 0.9986501019683699,
16 0.9772498680518208,
17 0.8413447460685429,
18 0.5,
19 0.15865525393145707,
20 0.022750131948179195,
21 0.0013498980316300933
22 ],
23 "monotone_decreasing": true,
24 "q_at_m_zero": 0.5,
25 "predicted_q_at_m_zero": 0.5
26 },
27 "covariance_sweep": {
28 "0.5": {
29 "s_h": [
30 0.0425,
31 0.050625,
32 0.05265625,
33 0.0531640625,
34 0.053291015625,
35 0.05332275390625,
36 0.0533306884765625,
37 0.053332672119140626,
38 0.05333316802978516,
39 0.05333329200744629,
40 0.053333323001861575,
41 0.053333330750465396,
42 0.05333333268761635,
43 0.053333333171904085,
44 0.05333333329297602,
45 0.053333333323244005,
46 0.053333333330811,
47 0.05333333333270275,
48 0.05333333333317569,
49 0.053333333333293924
50 ],
51 "ratio_last_first": 1.2549019607833864,
52 "stable_limit_if_applicable": 0.05333333333333334
53 },
54 "0.9": {
55 "s_h": [
56 0.048100000000000004,
57 0.078961,
58 0.10395841,
59 0.1242063121,
60 0.140607112801,
61 0.15389176136881,
62 0.1646523267087361,
63 0.17336838463407628,
64 0.1804283915536018,
65 0.18614699715841748,
66 0.19077906769831818,
67 0.19453104483563774,
68 0.1975701463168666,
69 0.20003181851666196,
70 0.20202577299849622,
71 0.20364087612878196,
72 0.2049491096643134,
73 0.20600877882809387,
74 0.20686711085075607,
75 0.20756235978911244
76 ],
77 "ratio_last_first": 4.315225775241423,
78 "stable_limit_if_applicable": 0.21052631578947376
79 },
80 "1.0": {
81 "s_h": [
82 0.05,
83 0.09,
84 0.13,
85 0.17,
86 0.21000000000000002,
87 0.25,
88 0.29,
89 0.32999999999999996,
90 0.36999999999999994,
91 0.4099999999999999,
92 0.4499999999999999,
93 0.4899999999999999,
94 0.5299999999999999,
95 0.57,
96 0.61,
97 0.65,
98 0.6900000000000001,
99 0.7300000000000001,
100 0.7700000000000001,
101 0.8100000000000002
102 ],
103 "ratio_last_first": 16.200000000000003,
104 "stable_limit_if_applicable": null
105 },
106 "1.1": {
107 "s_h": [
108 0.0521,
109 0.10304100000000002,
110 0.16467961000000006,
111 0.23926232810000012,
112 0.32950741700100017,
113 0.43870397457121024,
114 0.5708318092311645,
115 0.7307064891697092,
116 0.9241548518953483,
117 1.1582273707933717,
118 1.44145511865998,
119 1.784160693578576,
120 2.1988344392300774,
121 2.700589671468394,
122 3.307713502476757,
123 4.042333337996877,
124 4.9312233389762214,
125 6.006780240161229,
126 7.308204090595088,
127 8.882926949620057
128 ],
129 "ratio_last_first": 170.49763818848479,
130 "stable_limit_if_applicable": null
131 },
132 "1.4": {
133 "s_h": [
134 0.0596,
135 0.15681599999999998,
136 0.3473593599999999,
137 0.7208243455999997,
138 1.4528157173759995,
139 2.8875188060569585,
140 5.699536859871638,
141 11.211092245348409,
142 22.01374080088288,
143 43.18693196973044,
144 84.68638666067166,
145 166.0253178549164,
146 325.44962299563616,
147 637.9212610714468,
148 1250.3656717000356,
149 2450.7567165320693,
150 4803.523164402855,
151 9414.945402229596,
152 18453.332988370006,
153 36168.57265720521
154 ],
155 "ratio_last_first": 606855.2459262619,
156 "stable_limit_if_applicable": null
157 }
158 },
159 "covariance_predictions": {
160 "stable_lambda_0.5_limit": 0.05333333333333334,
161 "observed_lambda_0.5_final": 0.053333333333293924,
162 "stable_absolute_error": 3.941291737419306e-14,
163 "unstable_lambda_1.4_ratio_last_first": 606855.2459262619,
164 "predicted_unstable_ratio_gt_1": true
165 },
166 "horizon_sweep": {
167 "q_per_step": 0.1,
168 "horizons": [
169 1,
170 2,
171 3,
172 4,
173 5,
174 6,
175 7,
176 8,
177 9,
178 10
179 ],
180 "union_risk": [
181 0.09999999999999998,
182 0.18999999999999995,
183 0.2709999999999999,
184 0.3439,
185 0.40950999999999993,
186 0.46855899999999995,
187 0.5217030999999999,
188 0.5695327899999999,
189 0.6125795109999999,
190 0.6513215599
191 ],
192 "risk_at_h5": 0.40950999999999993,
193 "predicted_h5": 0.40950999999999993
194 }
195 },
196 "mini_experiment": {
197 "n": 2500,
198 "horizon": 8,
199 "violation_rate": 0.5168,
200 "reachable_risk_recall_at_5pct_fpr": 0.7097523219814241,
201 "mean_score_recall_at_5pct_fpr": 0.7329721362229102,
202 "reachable_threshold": 0.9403956062895902,
203 "mean_threshold": 0.9837043382706925,
204 "reachable_risk_ece": 0.11692485162830053,
205 "mean_score_ece": 0.08122748632856695,
206 "baseline_definition": "fixed-sigma endpoint Gaussian warning probability (sigma=0.05)"
207 }
208}