Classic arguments for decentralization, augmented by ideas about how participation empowers the poor, motivate the widely used approach in foreign aid called community-driven development. CDD devolves control over the selection, implementation and financial management of public goods to communities. Until recently, policy enthusiasm has outstripped the evidence. I synthesize findings from randomized controlled trials and find that CDD effectively delivers public goods and modest economic returns at low cost in difficult environments. There is no evidence, however, that CDD transforms local decision-making or empowers the poor in any enduring way. Part of this failure may be because some constraints believed important—like insufficient social capital—appear not to bind. Others, like exclusive local institutions, are a problem, however not one that CDD remedies. These results present a conundrum: how much participation is enough to safeguard the gains of such “extreme” decentralization, while minimizing the opportunity costs imposed on poor people’s time?