More granularity is still valuable though, since in topics where deep understanding is needed, the shallower levels of understanding are required to master the deeper. E.g., you need to be able recognize an integral before you can solve them. Which in turn you need to be able to do in canned rote mechanical cases before you can handle general word problems that involve assembling and solving integrals. And not all subsequent classes or careers will necessarily require the full mastery of the topics in a class. Plus it's good to have such detailed feedback for both the student and the professor to focus on shortcomings.

Another issue with the purportedly "optimal" testing situation of creating unique challenging and deep questions for each test (and even for each student). It's difficult to make good questions. Sometimes they are too hard, are not as clear as they seem since students might misunderstand and go off in the wrong direction giving you some trivial answer to a similar-looking question. The safest questions are the most shallow. And deeper problem-solving questions may take a few iterations of "testing" (as in trying them out on exams and seeing how students do). So what seems like the best kind of question will actually be the riskiest and least-tested.