There's a common problem (https://en.wikipedia.org/wiki/Goodharts_law) with proxy measurements, and issues with testing highlight this problem. However your description of what the math education looks like I think this is a lesser evil, at least until we find better measurable proxy for actual education that tests.