Public documentation

Scanner methodology

How scanner sources, provenance, confidence, and review categories are assigned.

Scanner sources

axe-core findings

Generated by the bundled local axe-core engine. A11Y Cat preserves rule ID, impact, target, help URL, tags, checks, and evidence where available.

A11Y Cat custom checks

Product-owned checks cover metadata, links, language, spelling, headings, alt/name analysis, contrast review, state coverage, diagnostics, review categories, local history, and export shaping.

Manual and review items

Generated when automation cannot safely confirm a failure, including incomplete engine results, visual ambiguity, hidden-state limits, link-purpose judgement, language heuristics, and assistive-technology workflow review.

Diagnostics and limitations

Explain runtime errors, inaccessible frames, restricted pages, skipped checks, suppressed areas, unsupported contexts, and export support boundaries.

Classification rules

OutputRequired confidenceTypical sourceReporting rule
Confirmed issueHighaxe-core violation or strict custom rule that passes reliability gatesCan appear in main scanner results as a remediation candidate.
Needs reviewHuman judgement requiredincomplete result, visual ambiguity, interaction dependency, editorial judgementMust be reviewed before being filed as a defect.
AdvisorySupportive, not defect-grademetadata, spelling, language specificity, quality signalUse as guidance or backlog input.
LimitationBoundary is knownrestricted context, missing state, frame boundary, unsupported evidenceRecord the gap and use another method.
DiagnosticOperationalpermission, storage, runtime, export, scan coverageUse for troubleshooting, not page-failure reporting.

Contrast methodology boundary

Confirmed contrast findings are shown in the main scanner only for axe-core color-contrast violations or custom rendered-text findings that pass strict reliability gates. Gradients, background images, transparency, overlays, dynamic backgrounds, CSS effects, text over media, known injected overlays such as BugHerd, disabled/inactive states, decorative/logo text, icon-like SVG content, image/canvas-rendered text, clipped/offscreen paint, unknown layering, and fallback-to-white cases are excluded, downgraded, or routed to review with a reason.

Design screenshot evidence

Image-only design review can discuss visible layout, hierarchy, spacing, contrast concerns, content clarity, and likely accessibility risks. It must not create fake DOM, selector, focus order, ARIA, accessible-name, semantic HTML, tab order, form label, or real screen reader findings from a PNG, JPG, or WebP screenshot.

Screen Reader Review evidence

The Screen Reader Review aid can organise expected reading order, visible content checks, review notes, virtual review output, and exports. It does not prove real NVDA, JAWS, VoiceOver, TalkBack, or Narrator behaviour unless the reviewer adds real evidence from those tools.