Built to pass the human, not punish them.

Keyboard navigation: Every interactive element reachable with Tab, activatable with Enter or Space. Focus indicators visible with a 3:1 contrast ratio against the surrounding background.

Screen readers: ARIA roles and labels on all controls. Live regions for status updates ("Verifying…", "Verified"). Tested with NVDA + Firefox and VoiceOver + Safari.

Audio fallback: One-click switch to audio mode from any visual challenge. Audio plays a synthesized phrase; user types what they hear. Lenient matching (Levenshtein distance ≤ 2 for words).

Accessibility-only mode: Sites can opt into audio-as-default for users behind their auth. No visual challenge ever appears for that site's traffic.

Reduced motion: Respects prefers-reduced-motion. Animations don't trigger on the success state when this is set.

High contrast: Respects prefers-contrast: more by raising stroke weights and removing soft borders.

Color contrast: All text-on-background combinations exceed 4.5:1 (AA Normal text), most exceed 7:1 (AAA). Tested with axe-core and Pa11y.

Form labels: Every input is paired with a programmatically associated <label>. No placeholder-as-label.

We don't have an "accessibility overlay." Those are a known anti-pattern that often makes things worse. The widget is accessible at the markup level.

We don't ship a "VPAT" — yet. We're a small team and a full VPAT requires audit time we haven't budgeted. We're working on it.

We don't claim WCAG 2.2 Level AAA — some criteria (e.g. extended audio descriptions, sign language interpretation of audio challenges) aren't applicable, and others would compromise the security model.