My apologies, I discovered the climb problem with circles myself recently. That will be fixed next release. (The toolpath generation for circles didn't seem to get updated after the milling direction logic was added). But the other problem is new to me.
In your .cb files you had both ops to be climb, so I set the 1st to be conventional then reran the g-code and did indeed see what you described.
What I notice is that for some reason when climb milling the toolpath starts at the
second point last point in the polyline rather than the first. So when you do a conventional followed by a climb (or vice versa) it does indeed jump between different starting points.
I will look into this and make sure climb milling also starts from point 1.
One other point to make, there is a difference between your CAD circle to polyline logic and CamBam's. I was originally breaking circles into 2 180 degree arcs (like your CAD), but this would occaisionally cause arc problems. If you have 2 points and a bulge defining the arc, it can be ambiguous which direction the arc goes. For 180 degree arcs, it just takes one rounding error in the wrong direction and the arc will go the other way. To play it safe I now use 3 arc segments for circles.
Of course I had to discover this strange behavior as I was bragging it up and demo'ing it to a client. Sigh. 
Reminds me of the time we had just put in a new server rack at an office in London. To demonstrate the robustness of the UPS system to the clients (MD, head of finance etc), my boss (with a theatrical flourish) unplugged the mains. You can guess the rest.
More great feedback Walt, thanks again!
