Here's some photos of how these bearings were used on 228's 2010 robots.
Here's our swerve module; we had assembled nine of these: four for competition robot, four for practice robot, one spare/pit display demo.
The internal gearbox had a 10.204:1 ratio (about 9.5 ft/sec normal) accomplished by a first stage 50:14 reduction (with a 28t idler to act as a spacer), and a 40:14 second stage reduction. The first stage reduction was accomplished between the inner CIM mounting plate and the outer U-shaped gearbox housing plate. The second stage reduction was done on the inside, between the CIM mounting plate and the Colson wheel. To get the torque through the cluster shaft from the first stage on one side of the CIM mounting plate to the second stage on the other, we used one FR6ZZ-Hex bearing.
After about about forty minutes of use, this is what a typical failure looked like.
To fix this issue, we disassembled the gearboxes and modified the cluster shafts. We left the hex on the inside to keep the shaft sandwiched between the two flanged FR6ZZ bearings, but turned the shaft from the bearing outward into round and milled a 3/32" keyway into it. This allowed us to retain the 50t 3/8" hex bore gear from the first stage, but excluded any use of the 14t 3/8" hex pinions for the second stage. Luckily, AM uses 14t gears for standard CIM pinions; we took four of these, bored them out from 8mm to 3/8", and broached them for a 3/32" key. After this, we've had no bearing problems of any kind.