Multiple things can be true, because the goal is to optimize in aggregate.

- Some teachers are bad (and some students will have them)

- Overriding teachers with policies intended to control the bad ones impairs and burns out the others

Consequently, the reasonable path is somewhere in the middle. Create feedback systems designed to identify and weed out the worse teachers* and avoid overloading everyone else with outcome-less proscriptive policies.

* F.ex. it consistently amazes me that few systems, teaching included, regularly poll their end users (students or employees). "Well, people will give bad reviews if they get bad grades!" No shit, and somehow that's something we can't adjust for with a basic statistical analysis?