In this paper, we show that the design and decentralized, school-based scoring of New York's high school exit exams--the Regents Examinations--led to the systematic manipulation of test sores just below important proficiency cutoffs. Our estimates suggest that teachers inflate approximately 40 percent of test scores near the proficiency cutoffs. Teachers are more likely to inflate the scores of high-achieving students on the margin, but low-achieving students benefit more from manipulation in aggregate due to the greater density of these students near the proficiency cutoffs. Exploiting a series of reforms that eliminated score manipulation, we find that inflating a student's score to fall just above a cutoff increases his or her probability of graduating from high school by 27 percent. These results have important implications for educational attainment of marginal high school graduates. For example, we estimate that the black-white graduation gap is about 5 percent larger in the absence of test score manipulation