Poorly controlled runout has haunted this project since first musings about its silly predecessor. Pulling back from an earlier somewhat helpful idea seems to have helped to escape a local maximum. At least for parts that can be finished with a single tool.

To prove the pudding, an even smaller differential gear set   ̶i̶n̶ ̶p̶r̶o̶g̶r̶e̶s̶s̶.̶.̶.̶   done:

I prolly should add something here about reducing runout with cheap dremeloid clones...

Already kinda glossed the method in this log. The main thing since then is just getting better at getting more predictable results with less effort. Also I partially undid an earlier attempt to constrain the poorly constrained collet, which made it easier to just push it around until it points in the right direction. 

The "in progress" pic:

Done with a cheap generic DremelⓇ-like rotary tool.