This paper describes the results of a corpus study of English auxiliary contraction and their implications for the place of variation in a grammatical architecture. Specifically, I argue that the quantitative patterns displayed by this variable lend themselves to a two-stage analysis of contraction, as follows: a variable process of morphosyntactic adjunction conditions insertion of either a full or a contracted form; then, low-level phonetic and phonological processes alter the phonological shape of that form. I support this analysis with data showing an effect of subject length on contracted forms, such that contracted forms are disfavored after longer subjects, with an apparent eight-word cut-off after which contracted forms categorically do not surface. I conclude by addressing the implications of this finding for a grammatical architecture, proposing that this eight-word cut-off may be best localized in a system of "usage" distinct from the grammar.