Vague predicates are obligatorily tolerant (Wright 1973).  For example, the positive form gradable adjective ‘long’ cannot be used to draw a sharp distinction, even when the facts of the context of utterance and the semantic properties of the sentence in which it occurs otherwise conspire to make such uses possible:  in a context in which there are exactly two salient objects A and B, which are clearly but only slightly different in length, it is infelicitous to use “the long one” to refer to the longer of the two; we must say “the longer one.” This despite the fact that the presuppositions of the definite article and the lexical semantic features of ‘long’ that underwrite a compositional account of the comparative ought to allow for such a use. In this talk, I will do three things.  First, I will argue that no existing account of vagueness — neither epistemic nor expressivist nor contextual nor semantic nor cognitive nor distributional — provides a satisfactory explanation for why tolerance persists in examples like the one above.  Second, I will show that attempts to eliminate the challenge presented by these kinds of examples through appeal to a grammatical decomposition of positive form adjectives into a non-vague root and a vague “pos” morpheme are also unsatisfactory, because they fail to generalize to other kinds of examples that manifest exactly the same behavior but cannot plausibly be analyzed in this way.  And finally, I will outline an alternative, pragmatic account of vagueness, in which  application of a vague predicate constitutes a “vote” for a particular resolution of linguistic indeterminacy, and tolerance emerges from a normative constraint treating such votes as subject to principles of social choice. This account is consistent with a transparent compositional relation between positive and comparative (in which the latter is derived from the former), and generalizes to non-adjectival cases of vagueness as well.