It is one thing to produce an innovative, construct-based assessment task; it's another to produce 10 a year that are comparable in difficulty, measure the same competencies, are free of differential item functioning, and can be scaled and equated. These challenges contributed to the failure of the performance (or authentic) assessment movement of the 1990s to gain much traction for summative uses. Many of those performance assessments were innovative and construct relevant, but failed to meet some fundamental requirements, including efficiency (in terms of cost and time) and psychometric quality (particularly in terms of reliability/generalizability, and form-to-form comparability). How can replicating these problems in the new science assessments be avoided? As Alonzo and Ke (2016, see EJ1122509) suggest, new science assessments will be needed to satisfy the new science standards, but the authors of this commentary suggest that new architectures for these new assessments will probably also be needed. Herein they discuss the uses of assessments, using embedded assessments for certification, and propose linking embedded assessments to standard certification assessments.