Gear reduction gives you slop, if you need to reduce the better way is probably using a kevlar toothed belt and two pulleys. Still not perfect but better than gears. Finally, the best way (and most expensive way) is by using a properly pre-tensioned ball bearing driven spindle ('ball screw').
I work in the non-destructive testing industry, testing aerospace parts. We build CNC testing rigs fairly regularly, with x and y axes longer than 30 feet, and carry loads of >1000 lbs. These need to be able to center on holes with diameters of <1 mm, because that's what we calibrate our equipment on.
The most reliable build setups we have are kevlar belts coupled with 16:1 reduced stepper motors, and decoupled encoders that index using constant pressure rack and pinions. Even pre-tensioned ball screws give us enough slop to be a problem without encoders, and kevlar belts are an order of magnitude cheaper; going with a reduced stepper vs a servo takes the cost down by half.
Ah good point, yes, at those lengths you really don't want ballscrews, they are impossible to support and will warp. But for short distance (up to 8' or so) with a support on either end they are fine.
Yeah, my brain is kinda tuned to those long distances. I also kind of harp on people who say they need really high precision systems, so use servos and spend bookoo bucks on slop-reductive hardware.
Even a cheap encoder, when decoupled from the motor, and directly coupled to the axis it measures, will give better axial positioning than a servo. It boggles my mind why people seem to ignore this in favor of a motor that knows where it is in space, but doesn't know where the axis it is driving is in space.
