The majority of U.S. public school districts now offer school choice programs that allocate seats using a centralized algorithm but with voluntary participation. The optional nature of public school choice segments public education and raises critical questions: who chooses, who benefits, and what policy alternatives can produce better outcomes for children? This paper provides new evidence on these questions by studying the Los Angeles Unified School District, the largest opt-in system in the country. Analyzing two decades of lottery records, we find that students living closer to choice options are both more likely to participate and to experience larger achievement gains. The proximity-based treatment effect heterogeneity is not explained by other observable treatment effect heterogeneity, suggesting a potential role for unobserved demand-side factors. To assess the empirical relevance of this hypothesis, we rely on quasi-experimental variation in distance to schools due to large expansions in choice programming and lottery-based admission lotteries to estimate a generalized Roy model that links families' decisions to apply and enroll to achievement gains. Our estimates indicate that the families least likely to apply under the current system would realize the largest test score benefits if they participated. In other words, decentralized, opt-in systems not only segment public education markets based on student ability and socioeconomic status but also exacerbate existing inequalities in educational outcomes. Policy interventions---such as targeted information interventions that cultivate broader participation or mandate participation as is done in cities such as New York---could produce sizable achievement benefits for school districts.