In one of my files I am working with a tolerance of 0.00001 instead of 0.0001 for several reasons that are not relevant to this observation.
When I PROJECT a curve onto a surface SPLIT fails at the lower tollerance. Just to be sure I EXTEND the curve to ensure full coverage of the surface but the SPLIT operation fails and the input surface is returned by the operation. If I SPLIT with a surface at this tollerance, it succeeds.
Some people here will say that fillets are bad types of surfaces to work with, because they are degree 2 with point weights. You will find that rebuilding that surface (to degree 5 with 10 CP - rebuilding the surface within 1e-8 tolerance) will stop the split from failing. It feels to be a bug though.
Aaaah ha. I suspect as much. Thank you for confirming that I need to rebuid that surface. I came to the same conclusion, and have begun to work on rebuilding every trimmed surface from the underlying geometry