This is a tongue-in-cheek rendering of an imaginary linguist's attempt to prove that a grammatical rule is not in fact optional, but conditioned. Through manipulation of the data, use of computers, etc., he succeeds in "proving" this. He also "proves" that this conditioning is contextual; that the rule which exhibits it is essentially probabilistic; and that hyperbolic tan functions are a feature of the competence underlying the human language faculty. An alternative, much simpler hypothesis is presented, but rejected on the grounds that, although it is compatible with the data, it is not "empirical" enough. Assumptions implicit in variable rule descriptions (VRD's) are examined, with the conclusion that the VRD approach has unfortunate methodological comsequences for linguistic theory. It is further concluded that probabilistic grammar is not a theory. (Author/AM)