Certain types of phonological opacity have proven to be problematic for strictly parallel versions of Optimality Theory (OT; Prince and Smolensky 1993). In this paper, I discuss one such case from Polish, an alternation between Zå\and Zt\, that is opaquely obscured by word-final obstruent devoicing. In § 1, I present the relevant data and generalizations, and I provide a skeletal OT analysis which demonstrates why these data cannot be analyzed in strictly parallel OT if both governing generalizations are synchronically productive. In § 2, I bring to light lexical and experimental data which suggests that the Zå\~ Zt\alternation is not in fact synchronically productive, clearing the way for a strictly parallel synchronic phonology. In order to account for the existence of opacity in the lexicon, I construct a diachronic analysis of these data in § 3 within a system of serially ordered parallel phonologies. I propose that a strong version of Prince and Smolensky‘s (1993) Lexicon Optimization encodes the results of each historical sound change directly into the evolving lexicon. Finally, in § 4, I summarize the major claims of this paper and pose questions for further study.