Because many of the forms participating in inherent variability are not attested in the standard language, they are often construed as evidence of change. We test this assumption by confronting the standard, as instantiated by a unique corpus covering five centuries of French grammatical injunctions, with data on the evolution of spontaneous speech over an apparent-time span of 119 years. Reasoning that forms salient enough to have attracted the attention of grammarians were likely widespread in the speech of the time, we demonstrate how these materials may be used to (i) infer the existence of prior variability, (ii) trace the evolution of normative dictates associated with the variants, and most revealing, (iii) discern hints of prior linguistic conditioning of variant selection. These are then operationalized as factors in a multivariate analysis and tested against the facts of usage. The linguistic focus is on future temporal reference, a notoriously variable sector of the grammar in which competing exponents have persisted for centuries. Systematic comparison of grammatical treatments with actual speaker behavior shows virtually no correspondence between the motivations offered in the literature and those constraining actual variant choice. Prescriptive efforts to explain variability, by ascribing to each variant form a dedicated reading or context of use, have had no effect on speech, which is shown to be governed by a powerful set of tacit variable constraints. These in turn are unacknowledged by the grammatical tradition. The result is a great and growing disconnect between the variable rules governing speech and the normative dictates that underlie the notion of the standard. We explore the implications of these findings for the use of grammarians' observations as data for linguistic analysis.