p, li { white-space: pre-wrap; } We study relational contracts between a principal and an agent when they face symmetric uncertainty about the match quality. Actions affect learning about the match quality and the principal's payoffs. Because the agent's actions are perfectlyobservable, the agent cannot bias the principal's beliefs. We show that even when the agent is not protected by limited liability and despite the absence of private information and hidden action, uncertainty about match quality precludes efficiency. The source of inefficiency is the holdup problem arising out of the separation betweenthe entity exerting effort and the entity collecting the output. We characterize the set of all subgame perfect equilibria of the associated game. We show that Pareto Optimal equilibria may involve actions that are dominated in their informational content as well as payoff. Such actions are a modest way for the principal to provide incentives and learn about the match quality, when more efficient ways are not credible. Conditional upon strong performance, we show that the relationships move to a phase where actions that offer better learning and higher payoff are used. In this phase the agent is rewarded with a bonus upon strong performance.