On this point in particular, most of the routes we've tried to use to quantify teacher worth have resulted in bizarre scenarios where teachers actively want to avoid certain classes of students. Early metrics usually punished teachers that were assigned underperformaners - while no child left behind had an expected grade growth formula that could result in a teacher needing to get a student a score above 100 to avoid being weighted towards getting a pay cut.

It's honestly a lot like development - it's hard to distinguish someone being lazy or bad from someone who's gotten stuck with difficult problems - or, more likely, problems that looked easier on the surface than they were in actuality. But, unlike development where these measures could potentially be discovered by reducing efficiency and forcing multiple devs to solve problems in parallel - that's impossible to do when every student is different at a basic level.