I have trouble buying in to a methodology that doesn't somehow stratify score ranges & run the analysis independently within those bins. e.g. You can drill the permutation formula into a kid w/ a natural 600 math score. A 450 score, who knows?
another thought here is that past a certain pt diminished returns to scale are driven by time-pressure (ie teaching ppl 1M techniques to try is counterproductive) so imbalanced skillsets might affect teachability