Since the recent changes pcucka made about six months ago to improve the OpenVDB Scatter SOP and add the "SDF Domain" options, I've noticed a couple of inconsistencies. 

The following tests use a Platonic SOP with default options except "radius/size" set to 5. I'm then using a Delete SOP to extract just points within a small slice (0.1) in one axis for visualization purposes.

1). With isosurface mode and points per voxel, there's the occasional bleeding of points away from the isosurface:

However, if I use a wrangle to sample the levelset and cull points in a narrow band from the isosurface, I can remove the offending points, so it doesn't appear to be a sampling issue:

This happens with both Houdini points and VDB points.

2). When scattering VDB Points, you also get the additional option to "Clip To Isosurface", which is valid for "interior" and "isosurface" modes. However, if you set the spread to 1 and enable "Clip To Isosurface", you only get points clipped on one side of the isosurface. This seems like an odd behavior to allow:

I suggest we only enable "Clip To Isosurface" when in "interior" mode and instead recommend the spread option to be used to force the points onto the isosurface just as it is for Houdini points.