The problem then seems to be tests that are too abstract. The students are optimising for tests that aren't requiring them to solve real-world problems.