A two-stage experiment disentangles the effect of various aspects of pay-for-performance contracts. The first is a lab-in-the-field experiment where 1,359 health workers are primed with a checklist of salient clinical tasks, then randomized within 690 clinics to receive no incentives, rewards, or penalties for treating hypothetical patients. Both rewards and penalties improve performance by 20 percent and generate spillovers on unincentivized tasks, but small incentives capture most gains. In the second stage, lab impacts translate into the real world: lab PFP exposure improves by 20 percent the care provided to real-world patients even after the lab experiment.