Works on My Machine

Accessibility bugs are not always visible in a screenshot or detectable through static checks. Sometimes a dialog closes, keyboard focus disappears, and a screen-reader user loses their place. Reproducing these failures requires connecting user interactions, spoken feedback, focus transitions, and source code. **Tabwise is an AI-assisted accessibility debugging and repair workflow built around what users actually experience.** Developers can record and save keyboard journeys, inspect screen-reader observations, identify failures, review source-level repairs, and compare the original behavior with a repaired rerun. Tabwise combines a React/Next.js workspace, genuine Windows NVDA capture, structured evidence, Meta Muse Spark reasoning, human-controlled patch review, and deterministic verification. Evidence and repair artifacts are source-bound and preserved for inspection. In our measured Fernway seat-selection example, dismissing a dialog incorrectly moved focus to the document body. Meta Muse Spark connected the recorded behavior to the dialog-close handler and proposed a targeted correction. After reviewing and approving the exact patch, we executed a fresh native NVDA run. The deterministic comparison reported **FIXED**: focus returned to the seat-selection button, and the tested surrounding keyboard and speech behavior was preserved. Tabwise turns accessibility debugging from guesswork into an evidence-driven workflow where a proposed fix can be inspected and measured.