I propose a model of democratic accountability (a la Maskin and Tirole, 2004) under supranational policy constraints. I derive three broad categories of perfect Bayesian equilibria (PBEs), namely polarized, technocratic, and populist equilibria, and show how the equilibrium reelection rule may change under different assumptions about the information structure, i.e., depending on whether voters are only able to observe policy outcomes, inputs, and/or outputs. I find that in contexts where governments are perceived to have little control over implemented policy and economic performance, i.e., in a "systemic" environment, voters do not look to outcomes but rather inputs in deciding whether to reelect a "responsive" (congruent) incumbent (i.e., "input legitimacy"). The observation of the incumbent's input in itself provides voters with information about the incumbent’s type irrespective of policy outputs and outcomes. On the other hand, when the underlying policy environment is perceived to be "endemic," voters will reward "responsible" (competent) incumbents (i.e., "output legitimacy"). Therefore, voters will tend to focus on inputs in systemically constrained contexts and outcomes in more unconstrained policy environments.