Can we use behavioral experiments to draw inferences about the evolution of human social behavior, and if so how? To illustrate the complexities, I examine the evolution of strategies for the repeated play of social dilemmas. Repeated interactions enjoy favorable status as an explanation for the evolution of human cooperation. So favorable, in fact, that one influential hypothesis paradoxically invokes repeated interactions to explain human cooperation in anonymous one-shot experiments. Some studies, however, suggest that repeated interactions in isolation, given a suitably flexible strategy space, do not reliably support cooperation. If generally true, this latter finding would contradict the status of repeated play as a basis for the evolution of both repeated and one-shot cooperation. Using a mix of analytical results and a large number of simulations, I consider this question with a social dilemma involving a continuous action space. Continuous action spaces arguably capture the social dilemmas people play in their daily lives better than games in which players can only choose to cooperate fully or defect fully. They also allow us to study the effects of extending the strategy space without positing super-human players who can remember long histories of past play. Under repeated play of such a game, I examine ten strategy spaces and seven approaches to implementation errors in all combinations. Results show an overall pattern of limited cooperation. In particular, given a combination of strategy space and implementation error that supports the evolution of cooperative strategies, some suitable generalization of the model typically destabilizes cooperation. Although these results do not constitute a general proof, they are consistent with the notion that repeated interactions only support the evolution of cooperation given arbitrary, biologically groundless restrictions on the set of strategies that can arise via mutation or innovation. This finding further implies that an important evolutionary explanation for one-shot cooperation enjoys a privileged status it does not deserve.