Often times matching input to output is already enough of a clue for the reader to understand the purpose of the functionality being tested. Often the code being tested is hard to use at all unless an example input is shown to the user. This is often my reason to read the unit tests: the function takes data of a shape that's very loosely defined, and no matter how I arrange those data, the code breaks with unhelpful error messages. This is, of course, also the problem of the function that I'm trying to run, but usually that's outside of the area of my responsibility, where my task is to just "make it work", not fix it. So, in this sense, unit test is a perfectly good tool to do the explanation.