Then include heterogeneity of approach in your ranking system. Even if we do this we still have a rank order from which we can select the N best and give them a

Then include heterogeneity of approach in your ranking system. Even if we do this we still have a rank order from which we can select the N best and give them a scalable platform through which they can reach large numbers, and be better off than under the archaic approach of a bijective mapping from random teacher X to Y students (Y less than 200), where the quality of X is probably dominated in a Pareto sense by any single teacher in the top N. Caveat here that I'm no expert in pedagogy, but I would also say that this fits well with the part of the teaching process that can be scaled with no downsides, eg lectures. But TA sessions etc can't be scaled as well and such an approach wouldn't fit.