Hi all,
Sorry for not being able to follow up on this issue earlier.
I believe Adam found the solution according to the last message.
Indeed, you need to use a surface that includes both the cortex and the cerebellum (let's label it S0) and then use it to generate the volume source space.
Before running the forward computation, you can check the placement of the sources, and you can overlay the FEM mesh on it and check if all the dipoles are well positioned.
If you find some dipoles outside the brain area, you can duplicate the surface S0 and then shrink the new surface by a few millimeters, then use this new surface S0' to generate the volume source.
PS: Please also note that the FEM may have some instabilities on the results when the dipoles are located on tissue interfaces with high conductivity gradients. This issue is still under deep investigation in the FEM modeling field.