This paper proposes a general way to craft public policy when there is no consensual account of the situation of interest. The design builds on a dual extension of the traditional theory of economic policy. It does not require a representative policymaker's utility function (as in the literature on ambiguity), a reference model (as in robust control theory) or some prior probability distribution over the set of supplied scenarios (as in the Bayesian model-averaging approach). The obtained policies are shown to be robust and simple in a precise and intuitive sense.