At the time of this reply, there are at least five competing definitions being argued for and against in this thread; some either requiring that the input and output be machine code or disallowing it, some necessitating that both input and output be at "the same level of abstraction" and others arguing that inputs and output can be at different levels of abstraction.

Given the large inconsistency and incompatibility between all these definitions and the existence of a well-accepted term that comprises all of these definitions, is it a mystery why people question the utility of the term? After all, what use is a nuanced term if it can't reliably convey that additional information.