IMO the correct way to design a test is to make it hard enough that the median score is around the 50% mark in order to capture the entire bell curve of performance, but then curve it enough afterward that everybody who met the standard the professor was actually trying to teach gets a good grade.