Model evaluation is still more art than science. New models claim to have superior performance every week. Practitioners have their own favorite models. Researchers continue to develop frameworks, only to have unique use cases break them.