Tools such as Alloy enable users to incrementally define, explore, verify, and diagnose specifications for complex systems. A critical component of these tools is a visualizer that lets users graphically explore generated models. As we show, however, a default visualizer that knows nothing about the domain can be unhelpful and can even actively violate presentational and cognitive principles. At the other extreme, full-blown custom visualization requires significant effort as well as knowledge that a tool user might not possess. Custom visualizations can also exhibit bad (even silent) failures.

This paper charts a middle ground between the extremes of default and fully-customizable visualization. We capture \emph{essential} domain information for \emph{lightweight} diagramming, embodying this in a language. To identify key elements of lightweight diagrams, we ground the language design in both the cognitive science research on diagrams and in a corpus of 58 custom visualizations. We distill from these sources a small set of orthogonal primitives, and use the primitives to guide a diagramming tool called Cope-and-Drag (CnD). We evaluate CnD using sample diagramming tasks and three different user studies, and find that short CnD specifications consistently improve model comprehension over the Alloy default. CnD thus defines a new point in the design space of diagramming tools: one that is lightweight, effective, and driven by sound principles.