Skip to main content

How Synli scans: our methodology

Synli builds on the open-source axe-core engine run through Playwright, then layers AI vision and Nordic-localized findings on top — and is honest about what automation can and cannot verify.

Accessibility tools are only useful if you can trust what they claim. This page explains exactly how Synli scans, which parts are automated, where human review is still required, and the regulatory details we encode so a report reflects the rules that actually apply to you.

Built on axe-core, extended by Synli

Synli's automated rule pass is powered by axe-core, the open-source accessibility engine (MPL-2.0) that also powers the axe DevTools used widely across the industry. We run it through Playwright so pages are evaluated in a real browser, including states a static parser would miss. Naming the engine matters: detection is a shared commodity layer, and we would rather be transparent than imply a proprietary black box.

On top of the engine, Synli adds the layers that are genuinely ours: AI vision analysis that judges whether an image's alt text describes the image, Nordic-localized rule output, authenticated scanning of pages behind a login, calibrated confidence scoring, and one-click transfer into the Norwegian uustatus.no accessibility statement.

The scanning pipeline
  1. CrawlPlaywright visits pages like a real user, including authenticated states
  2. Rule passaxe-core runs the deterministic WCAG checks
  3. AI layerVision and semantic analysis for judgement-heavy issues like alt-text quality
  4. Map & scoreFindings linked to WCAG criteria with a confidence level
  5. ReportVerified vs needs-review, exportable to uustatus.no or Markdown

What automation can and cannot verify

Industry studies and our own methodology put the share of WCAG issues a tool can reliably detect at roughly 30 to 40 percent. We do not inflate that. Synli marks every finding as either tool-verified or needs-human-review, so the report never overstates conformance.

  • Reliably automated: colour contrast ratios, missing alternative text attributes, form labels, language attributes, name/role/value exposure, and many ARIA misuse patterns.
  • Requires human judgement: whether alt text is meaningful, logical reading and focus order, link text in context, error-recovery quality, and whether content makes sense to assistive technology users.
  • Synli's AI layer narrows the gap on some judgement calls (notably alt-text quality) but is presented as assistance for a human reviewer, not a replacement.

The Norwegian 48/35 split, encoded

WCAG 2.1 has 78 testable success criteria. Norwegian law requires 35 of them (Level A and AA, excluding 1.2.3, 1.2.4 and 1.2.5) for the private sector, and 48 for the public sector — the extra 12 arriving with the EU Web Accessibility Directive, in force since 1 February 2023. Synli derives the applicable set from your organisation and solution type, so the report shows the rules that actually bind you rather than a generic checklist.

Synli does not replace manual accessibility review. It makes issues clearer, separates verified findings from items needing a human check, and structures the output so the remaining human work is targeted and fast.

Sources