It works when you deactivate the shell offsets and model the gaps between the sheets instead.
I noticed a similar shell-offset/contact limitation here, where a boundary condition includes nodes of different element sets, one of which includes the shell-offset and the other does not:
https://prepomax.discourse.group/t/limitation-or-bug-with-shell-offset-and-contact/990/8
Edit
Oh no, I noticed it’s exactly the same case… If you remove the corner nodes from the BCs (these nodes are shared by elements with different shell offset directions), the solution also converges with shell offset:
but obviously this leads to singularities at these locations
