Mesh-to-mesh constraints -- do they work? (bullet 2.82)

[yellowshuttle]

We have a dynamic btGimpactMesh constrained to a kinematic one. We're seeing a "pop" at the second frame of the simulation (i.e. the dynamic mesh moves in a physically-implausible manner from the "rest" position), after which the movement appears correct.

If we use a convex hull instead of a mesh, it works correctly (as does a box).

I've already made sure that the bullet rigid body is located at the center-of-mass of the mesh shape, and principal-axis-aligned.

Does this ring any bells? The mesh in question is concave -- should I be doing a convex decomposition myself?

Any help is appreciated.

-Arun