In "Updating the Duplex Design for Test-Based Accountability in the Twenty-First Century," Bejar and Graf (2010) propose extensions to the duplex design for large-scale assessment presented in Bock and Mislevy (1988). Examining the range of people who use assessment results--from students, teachers, administrators, curriculum designers, educational researchers, boards of education, to the public at large--Bock and Mislevy laid out a data-gathering design and analytic approach that would efficiently inform many of their needs. Bejar and Graf survey intervening developments that can be layered onto this design. They include cognitive frameworks for task development, real-time automated scoring for some types of performance tasks, and computer administration for multiple stage and adaptive administration, with blocks for various purposes such as linking time points and focusing on material that students are currently studying to provide formative feedback. This article presents the author's commentary on Bejar and Graf's (2010) article and focuses on designing assessment systems as an exercise in design under constraints. Bejar and Graf point out recent technical and scientific developments that help designers meet constraints in multipurpose large-scale assessment systems. Showing how these developments can extend the duplex design brings forward a coherent package of inferences, data collection designs, and analytic procedures from the original configuration. (Contains 1 footnote.)