1No matter how often we demonstrate it, remonstrate its neglect, or illustrate its terms and tropes, there’s no better way to teach close-reading skills than forcing students to try it. Individual results will vary, but the aggregate result will be a working consensus about the patterns and variations in a text that seem to reveal the writer’s deliberate choices, or have the greatest effect on readers. This paper offers a framework for assignments that use custom TEI schema to compile students’ annotations of a common text’s features (linguistic, semantic, structural, cultural) as a means of comparing human interpretations. It builds on the work of Kate Singer1 to overcome the limits of close-reading terminology with an expanded tagset. The paper addresses the interface required for effective crowdsourced markup, including example libraries among its training materials. It extends to humanities research the networked-science models (Michael Nielsen, David Weinberger) that leverage humans’ cognitive surplus (Clay Shirky). And it posits a machine-learning outcome of this pedagogical strategy: to build robust training sets to teach machines to identify low-level features, like rhetorical schemes (“Featured like him, like him with friends possessed”) or similes (“my state, Like to the lark at break of day”). Such automated detection might train human readers, while freeing them to concentrate on more complex high-level features like metaphors (“It is the star to every wandering bark”) or allusions. All is predicated on a set of interoperable close-readings that enable, but do not foreclose, new interpretations of texts at every level.