Hi Steve, it does appear to be related to very small differences in the tangents (like around 0.035 degrees), which get magnified to a certain extent during offset.

Do you happen to have any earlier versions of this before the curve filleting went in place?

Also, are you able to finish this model now, or do you need help tuning up this particular case?

I think that the fix will be to true up tangents that are really close before doing the offset. I'm also planning on similar tune-ups just in Join in general, and if I have a test case where the curve fillet is not accurate enough that would help to improve that as well.

- Michael