Two classic and previously unrelated problems are reconsidered for their implications for optimality theory and acquisition. The puzzle-puddle-pickle problem centers on the debate over children’s underlying representations and the characterization of interacting error patterns which, when lost, result in overgeneralizations. In response to the challenges that this problem poses, an optimality theoretic solution is offered that appeals to the second problem, the Duke-of-York gambit, which involves co-occurring generalizations with reverse effects. The solution avoids language-specific restrictions on input representations and characterizes the loss and introduction of errors by one mechanism. New insight is offered for when overgeneralization is (not) expected to occur.