Ten years ago, many policy makers viewed the reform of teacher evaluation as a highly promising mechanism to improve teacher effectiveness and student achievement. Recently, that enthusiasm has dimmed as the available evidence suggests the subsequent reforms had a mixed record of implementation and efficacy. Even in districts where there was evidence of efficacy, the early promise of teacher evaluation may not be sustainable as these systems mature and change. This study examines the evolving design of IMPACT, the teacher evaluation system in the District of Columbia Public Schools. We describe the recent changes to IMPACT, which include higher performance standards for lower-performing teachers and a reduced emphasis on value-added test scores. Descriptive evidence on the dynamics of teacher retention and performance under this redesigned system indicates that lower-performing teachers are particularly likely to either leave or improve. Corresponding causal evidence similarly indicates that imminent dismissal threats for persistently low-performing teachers increased both teacher attrition and the performance of returning teachers. These findings suggest that teacher evaluation can provide a sustained mechanism for improving the quality of teaching.