Found a bug in sift value. For me it seems that it takes only 2 correct epsilons for all input filters. Thus instead of 21 filter inputs and epsilons, I got only 2 epsilons for my 21 filter values.
bug_sift.v4p (12.7 kB)
hi frank, thx for reporting this. the sift node used the first epsilon only.
fixed in beta > 23.