Cleaned up a lot of the stuff hitting and a lot of the tolerances too. I also had to do some cooked surface modeling with onshape (I have never surface modeled before). Here’s what I did:

• The blue pieces were also hitting at the bottom, so I had to make the material even thinner than it was (really cooked, near 1.5mm thickness and barely clearing eachother). The standoff placements really cook everything so I’ll probably just extend the plate and move the standoffs further from the gears. However it will be uglier, and I would rather have a good looking cvt that doesn’t work than a ugly cvt that works.
• I had to do some really cooked stuff with surface modeling. Basically, the wrap feature wasn’t working. I spent like an hour or two trying to learn surface modeling, but nothing was working because the shape was weird, so I ended up just doing 8 separate revolutes, boolean, and then some move faces. Also, the bumps can’t be radial because the bearing isn’t radial (it faces horizontal, so the edge closer to the center moves up first which causes the whole thing to jam). To fix that I had to do even more cooked remove revolutes, then bridge curving and then knitting (I had never used these before) and fileting. I need to figure out a clearer way to do it because it’s getting on my nerves.
• I went through the latest V3 and compared around half of the holes to the V2 (it turns out there’s like 3 holes per part and theres like 100 holes to go through, not including clearances). The clearances also cooked some of the spacing so I went and fixed that.

I’ve been spending some time thinking about the 1way bearing solution, and I completely switched up and now think it’s impossible.