A characterization test (also called a golden-master or snapshot test) runs the code on many inputs, saves the outputs to a file, and later fails if any output differs. It does not claim the behavior is right, only that it is unchanged.