The GenScope Assessment Project is studying assessment in the context of a month-long computer-supported learning environment for introductory genetics. Across three annual iterations with multiple teachers, project researchers manipulated the materials, incentives, and contexts in which students were invited to use formative feedback on challenging classroom performance assessments. The consequences of these manipulations on engagement and learning were systematically examined from behavioral/empiricist, cognitive/rationalist, and situative/sociohistoric perspectives. This "comparative approach" was intended to provide new insights into unresolved issues over extrinsic rewards and accountability-oriented reforms.It turned out that the comparative approach also provided a powerful framework for refining and improving theories about classroom assessment. Essentially researchers "tuned" the classroom assessment environment to maximize gains on carefully aligned external performance assessments. Successive improvements led to correspondingly larger gains on an external achievement test that was more aligned with conventional genetics instruction. This study shows that design-based research around classroom assessment can help meet the wider educational goals of researchers within the increasingly narrow policies of reformers. Five appendixes contain examples of feedback and rubrics. (Contains 10 figures and 80 references.) (SLD)