Flow-A11y targets accessibility failures that appear only during interaction, such as keyboard traps, focus loss, modal leakage, delayed status updates, and dynamic controls. The system uses a real browser to execute natural-language scenarios, records ordered runtime traces, builds criterion-specific evidence packets, gates unsupported judgments, and produces auditable findings. On 19 public-web scenarios covering 45 dynamic WCAG criteria, it achieved more than ten times the oracle agreement of a generic browser-agent audit. Its evidence-calibration layer increased fail precision from 23.5% to 41.4% and removed invalid evidence references. The paper argues that runtime traces can support automation of accessibility checks traditionally requiring manual evaluation.
No heat snapshots are available in the last 24 hours.