I like the way you use the project cutter method to find the touching point of the surface.  As I was thinking about how to write a simple but versatile CAM plugin, I came across this treasure book: http://web.mit.edu/hyperbook/Patrikalakis-Maekawa-Cho/

I didn't have time to read it through yet, but it is definitely interesting.

Not sure, how to best do projection with Ball-end mills though.. 

I've been thinking about that problem.  In the Y axis, the issue is resolved by small enough stepovers as the material left by the cylindrical radius is cut by the following contour.  The X axis is a different story.  See sketch below.

The first option and perhaps the simplest to program, would be to do a second parallel surfacing perpendicular to the first.  In doing so the issue is resolved as in the Y axis, AND the scalloping from the first process is cleaned up.  

The second option, which I find interesting for projects that will only receive one finishing pass, would be to:

- check the normal angle of the point following the one being projected

- determine the point of contact on the cutter geometry (in this case a semicircle, viewed in elevation) with that normal

- evaluate the radius of the cutter at that angle

- project THAT radius at the point being cut.  

In this way the cutter tool is predicting the normal of the surface to come, by one step, and using a more appropriate tool radius for it's Z offset that a simple tool diameter.

I think for the time being I am going to focus on some other operations of the upcoming internship course, keeping this in the back of my mind as an alternative to multiple parallel surfacing passes for finished pieces.

Thanks for bringing it up, I had rationalized to myself that the stepovers resolved the whole problem, when they only did in the Y axis.