Automated audits prevent regressions but cannot certify accessibility

A perfect automated score can coexist with a checkout flow that disabled users cannot complete.

Tools reliably detect selected machine-testable failures, including missing names, invalid ARIA, labels, and some contrast problems.

W3C evaluation guidance states that no tool alone determines conformance because knowledgeable human evaluation remains necessary.

Automation belongs in CI because repeatable checks catch known regressions cheaply across frequent interface changes.

When alternatives require contextual judgment, keyboard tasks reveal behavioral failures unavailable to static rules.

Manual review should inspect focus order, screen reader output, error recovery, responsive states, zoom, media, and complete user journeys.

Testing with disabled participants reveals barriers that both tools and nondisabled experts may systematically overlook.

Use automated results as a fast diagnostic layer, then judge accessibility through representative tasks, technologies, and users.