> Very little of what I used on a day-to-day basis came from rote learning The question is not how much non-rote learning is required to adequately perform any

> Very little of what I used on a day-to-day basis came from rote learning The question is not how much non-rote learning is required to adequately perform any task, but whether rote learning is a necessary component of it. By engaging with research papers (often requiring multiple readings), you "accidentally" also memorize a lot of the material (even if not for ever and ever), which makes it possible for you to "try to improve it" etc. So all I am saying is that rote learning is a necessary component of applying any suitably advanced material. Whether that happens organically by engaging with the material deeply enough or on purpose by simply rote-learning is orthogonal to the need for that knowledge to be there. At issue here in this thread is whether testing for such knowledge is signalling anything of value. I still believe it is because you'll have that knowledge with either approach, even if only one of those approaches will gain you more functional knowledge: basically, such a test is a proxy for you having engaged in the material. Which is why I suggest a mixed approach in a sibling comment: test for knowledge of "factoids" to simplify test development and application of such knowledge where you are required to manipulate "learned" concepts (so only part of the test is hard to create and keep consistently hard/easy).