Possible related - there is a weld function in the API that recalculates the normals for overlapping vertices but that might only be for within the same mesh and not span parts. It would interesting to look at this from the ZB side too to see if the issue can be avoided in the first place.