Chip load: fascinating, so I’ll definitely give the speed controller a try. Feed rate: wow, you were mowing through th e wood at twice the speed I am! Hmm.
The tool “popping out” looks like the thing is trying really hard to get through the wood, and when the wood finally stops resisting, it pops through. That makes me think the tool isn’t “cutting like butter”, which makes me wonder if the fault is in the choice of tool. Can you recommend a starter milling bit that you know shouldn’t be a problem? Given how easy it is to change a tool, I’d like to at least rule out the possibility that I’m battling a crappy lowest-bidder tool.
I did confirm that the spindle is turning the correct direction.
I wouldn’t necessarily say the rigid mount made it worse — those results were from single runs; there was enough variation there that the most I can say is that going to the rigid mount didn’t make anything better.
I will try your suggestions and pull the middle assembly apart, disconnect the belts, scrap the extra bits, make sure everything rolls smoooothly, and see how far that gets me.
In the meantime, I did a little deflection experiment to see if I could argue that the belt play was causing a problem. First I measured the tool deflection by placing a caliper between the tool tip and a clamped block (motors enabled), and pulling on the tool with a luggage scale until I got to 5 pounds force. I found that 5 lb deflected the tool by 1.2mm, and 2.5 pounds deflected it by 0.44mm (with a coefficient of variation of 5-20%). Then I moved the caliper up onto the rail to measure how far the roller assembly moved. The hypothesis was that the belts were to blame, so the roller must be moving the same amount as the tool (if the rest of the middle assembly were behaving rigidly). The roller deflected 0.50mm and 0.19mm (for 5 and 2.5 pound pulls, cv 7-10%). This suggests that the belts definitely aren’t the entire problem (they only explain about 40% of the deflection at the tool), and yet they’re also not entirely rigid.
Based on the deflection-at-the-tip experiments, my tool seems to be exerting lateral forces around 3.5 pounds. Maybe that’s ridiculously high (not “like butter”), and fixing the cutting tool would resolve the problem by reducing the force and therefore all sources of deflection. In any case, this experiment has convinced me that my ideas for removing flex from the belt assembly are probably not a useful way to spend time right now.