Computational models are presented that evaluate different theories of sound change, particularly with regard to the actuation of change. Standard phonologization of coarticulation models predict counterfactual across-the-board change (cf. Weinreich, Labov, and Herzog 1968). Models that simulate a sigmoidal trajectory of change are more empirically appealing, but also are very sensitive to initial conditions. It is proposed that herein lies the solution to the actuation riddle. Sound change arises when a linguistic leader (Labov 2001) perceives an incidental correlation of social and phonetic variables, and adopts her speech to the "change." This simple incident leads to an entire sound change. We expect sound change to arise with the same frequency as these spurious correlations. The (presumed) infrequency of such correlations offers a schematic solution to the actuation problem.