Accessibility testing methodology and limitations

Who publishes and maintains this guidance

Publisher and organizational author: ADAReporting. Source and implementation review date: September 26, 2026. This page describes the current reporting workflow and cites external standards separately from our product behavior. It is not an independent audit or a claim of professional certification.

An automated validation result is not a human screen-reader review. Our editorial policy explains how authorship, reviewer credits, and the scope of a review are identified.

HTML scanning and finding classification

The crawler tests supported rules against reachable public HTML pages. It checks response type before applying HTML rules and accounts for redirects and resources that could not be tested. A raw image, video, feed, or downloaded document is not treated as an HTML page.

Reports separate findings mapped to WCAG rules from best-practice recommendations and manual-review items. They retain page-level occurrences while grouping likely shared-template repair candidates. A repeated component can have many affected pages without requiring many independent code changes.

Available evidence includes the URL, rule, selector, captured element context, and inspection links. Evidence availability depends on the rendered page and scanner version. A failed evidence capture is not proof of an empty accessible name. A historical report is not silently enriched with newly observed evidence; targeted rechecks need their own date and scope.

PDF validation is a separate stage

Discovered PDF candidates are downloaded and validated with veraPDF against the applicable automated PDF/UA profile. The stage is bounded by document count, file size, execution time, and download limits. A report must state items that were not tested; a download or parser failure is not an accessibility failure.

A failed PDF/UA rule is a specific conformance result. Identification metadata, document structure, fonts, links, and content alternatives have different implications and should not be collapsed into a claim that every failed file is unusable. An automated pass still requires human review of reading order, meaningful alternatives, tables, and actual use.

The current worker configuration permits up to 200 document candidates per scan part, with a 20-minute document-stage budget, a 20 MB download limit, and per-document timeouts. Large or unavailable collections may require follow-up; the report records those limitations rather than claiming complete coverage.

What forms and media results mean

Finding a form does not establish that submission, error recovery, or confirmation works accessibly. A video inventory does not establish accurate captions, audio description, or accessible player controls. These require interaction and meaningful content review.

Documents, presentations, and videos are reported separately because each requires different accessibility checks. Counts describe discovered occurrences, not necessarily unique files or every state of an interactive widget. Changes between reports must be interpreted with the scope and URL inventory.

Human review and repair acceptance

Use keyboard-only navigation and an appropriate screen reader to complete important tasks. Check focus behavior, control names, instructions, error feedback, zoom/reflow, and dynamic states. Test documents in the tools and assistive technology used by the audience.

A repair should include a source change, a retest result, and evidence that the relevant task works. A report is an input to that process. We deliberately do not publish a numeric compliance score: absence of automatically detected findings is not a percentage of accessibility or a legal conclusion.

Report quality and corrections

Reports undergo automated PDF/UA validation during generation. This checks aspects of the PDF’s structure, but does not establish that its reading order or usability has been verified by a person using a screen reader.

A corrected report should explain what changed and whether your website was rescanned or the existing results were reprocessed. This distinction helps you tell a change in the report from a change to your website.