Skeleton tool visualizes navigation structures for accessible data viz

When sighted practitioners build accessible data visualizations, they currently code navigation structures (nodes, edges, input bindings for assistive tech) with zero visual feedback. This lack of representation prevents them from developing intuition about what makes navigation good or bad, capping the quality of non-visual experiences. A new research paper from Frank Elavsky and colleagues at CMU (accepted at IEEE VIS 2026) addresses this with Skeleton, a direct-manipulation authoring environment that makes navigation structures visible and editable.

Skeleton bundles three contributions: an Inspector that renders navigation graphs as interactive node-link diagrams, a Dimensions API that lets authors express navigation in terms of data dimensions instead of explicit graphs, and the Skeleton editor itself. Key techniques include a dual-view editor showing both the system's navigation model and the end user's spatial experience, a scaffolding engine that automates node placement by repurposing the visualization rendering pipeline, a live label-template editor with real-time screen-reader-output preview, and a testing mode where traversal sequences become visually trackable. In an in-situ study with 8 practitioners across visualization design, engineering, and research, making navigation structure visible changed how they engaged: they reconsidered their own visualization architectures, attended to more input modalities, and shifted from treating accessibility as a compliance task to treating it as a design problem.