Bug in conversion from diffstrata to tautclass

While doing some experiments, I stumbled upon the following inconsistency in computations:

sage: from admcycles import *
sage: X = diffstrata.generalisedstratum.GeneralisedStratum(sig_list = [diffstrata.sig.Signature(tuple([8,-3,-2,-3]))], res_cond = [[(0,1)],[(0,2)]])
sage: X.psi(1).evaluate()
9
sage: v = X.ONE.to_prodtautclass().pushforward()
sage: (v*psiclass(1,1,4)).evaluate()
27/2

I think that for X.psi the labels of markings run from 1 to n and I am sure about this for psiclass. So the above two numbers are not consistent. Both from theoretical results and a separate computation within admcycles I expect that the correct answer above is 9, which would mean that the problem happens with .to_prodtautclass().pushforward().

I am happy about any help from @jschwab or @jonathan.zachhuber to figure out what goes wrong here.