I'm encountering all the same issues in this thread, and agree with @EnokV that exposing all the contacts that a collider makes with a mesh is the way to go, rather than the current case of some aggregated/first found(?) vector that can't be readily picked apart. There's a definite loss of very handy data, which I'd love to see available to all. It's nice to simplify the api where possible, but in this case, the detail vs. the mesh is very absent. Passing back a list of triangle ids and a list of respective depenetration vectors would be pretty great. We'd have enough to fix it properly, hopefully once and for all! I feel like I've put at least a man-year into bodging around it, and never getting a solution which is robust or even satisfying.