We provide two methodological insights on \emph{ex ante} policy evaluation for macro models of economic development. First, we show that the problems of parameter instability and lack of behavioral constancy can be overcome by considering learning dynamics. Hence, instead of defining social constructs as fixed exogenous parameters, we represent them through stable functional relationships such as social norms. Second, we demonstrate how agent computing can be used for this purpose. By deploying a model of policy prioritization with endogenous government behavior, we estimate the performance of different policy regimes. We find that, while strictly adhering to policy recommendations increases efficiency, the nature of such recipes has a bigger effect. In other words, while it is true that lack of discipline is detrimental to prescription outcomes (a common defense of failed recommendations), it is more important that such prescriptions consider the systemic and adaptive nature of the policymaking process (something neglected by traditional technocratic advice).