The decisions states make regarding what their assessments look like and what kind of information they produce inevitably shape instruction. Since the No Child Left Behind Act of 2001 ushered in an era of testing-based accountability for schools, state assessments have been governed by a set of design decisions that emphasize easily generated, easily compared scores--even when these assessments are somewhat superficial proxies for the rich performance expectations state standards set for student learning. This makes sense if state assessments play a narrow and siloed role, focused on sending up a red flag around school performance and triggering a cascade of follow-up actions. While this might be consistent with how designers intend for assessments to be used, there have been unfortunate and unintended consequences for teaching and learning. Many states want to construct assessment systems more deliberately to achieve their goals. By leveraging our understanding of how various groups use information from state assessments, we can design assessments and systems that have a net positive impact on instruction. Based on evidence from assessment system design and implementation as well as lessons learned working alongside various states, a set of design principles emerge that govern assessments intended to support teaching and learning. By centering features of assessments that support better student learning experiences, teacher practice, and systematic supports and decision-making, we can create assessment systems that have a net positive impact on instruction. The design principles detailed in this report reflect ambitious but accomplishable goals for assessment systems--and large-scale systems, including states as well as national and international programs, are already on the path to making this work a reality. As systems move forward, keeping "positive instructional impact" as the North Star and centering decisions on specific instructional shifts from the current state of teaching and learning that assessments should support can help system designers make the best decisions within their local contexts.