Developing countries often face a cycle where weak tax compliance limits public goods, cutting incentives to pay taxes. We test whether improved local infrastructure can disrupt this cycle, using a randomized street paving experiment in Acayucan, Mexico. Of 56 eligible street projects, 28 were randomly selected. A model highlights two mechanisms: belief updating about government efficiency and reciprocity from direct benefits. Three implications follow: (1) belief updating occurs through exposure to paving anywhere in the network; (2) compliance rises with broader exposure; (3) reciprocity boosts compliance among directly treated owners. Survey data supports belief updating: among initially dissatisfied residents, a one-SD increase in exposure to assigned paving lowered dissatisfaction by 7.9 pp, while exposure to actual paving lowered it by 8.8 pp, with no effect among the satisfied. Property tax records show exposure to assigned paving raised compliance by 1.5 pp, and to actual paving by 2.6 pp (3% above baseline). Reciprocity mattered too: owners whose street was assigned paving (or actually paved) increased compliance by 3.2 pp (4.8 pp, or 5.5% above baseline). Belief updating yields four times as much revenue as reciprocity.