Replace scale_sig() with Scale_sig() to reduce regressions, as it may have better precision.