Weak classifier, strong rank test
Power at beta=0.95 · 100 trials per benchmark · N=256 · calibration m=20
Mean shift
C2ST
.81
Conformal
1.00
Covariance scale
C2ST
.00
Conformal
.47
Anisotropic
C2ST
.00
Conformal
.03
Heavy tailed
C2ST
.01
Conformal
.16
Additional mode
C2ST
.71
Conformal
1.00
Mode collapse
C2ST
.99
Conformal
1.00