Next, I added bearings to reduce friction and even added observation holes to monitor the setup. 

With vents, it worked temporarily, but failed while sealed. After dozens of iterations (and switching to stronger nylon filament), 

the differential finally survived long enough to move on.