That is odd, for 2 reasons:
1: this curve is already the result of a pull this is reflected in point 2:
2: the maximum deviation between this curve and the pulled is 6.25278e-13
Deviations of this magnitude should not yield such different output.
Regardless, do you agree this is a bug in the unroller that needs fixing, as the result is obviously faulty?
In our automated pipeline we process many-many curves like this so I don’t see a workaround option in pulling twice, especially since this might only help in this particular case and does not fix the root cause.