> Meticulous records real user sessions from a web app and replays them against each commit to catch visual regressions.
>
> Full AI-readable docs index (complete project setup journey + every page as markdown): https://app.meticulous.ai/llms.txt

# Accessibility

Meticulous runs an accessibility scan at every screenshot it compares during replay of your application, then reports the **new** accessibility regressions introduced by the change under test. Pre-existing issues never fail the check; only freshly introduced violations are surfaced.

![The accessibility check in the Meticulous UI: a failed check reporting new violations, each with its rule, severity, the failing element, fix guidance, and links to the sessions where it was seen](/docs/built-in-checks/accessibility-check-example.png)

## Verdicts and CI

By default, new **critical** and **serious** violations mark the Meticulous accessibility status check as failed on the pull request and require a reviewer to acknowledge them in Meticulous, while **moderate** and **minor** ones are reported as warnings that leave the check passing.

The check can also be made **informational only**: every new violation is then reported as a warning and the accessibility status check never fails, so you can see what it finds without blocking any pull request.

By default, a violation is treated as pre-existing if it appears in **any** session's base replay in the base run (not only the same session's). That avoids reporting long-standing issues as new when a head replay reaches a screen its own base replay did not. You can turn this off in project settings to require a same-session match.

## Why this complements static analysis

Static analyzers and linters inspect your source code, so they only see accessibility issues that are visible in the markup you wrote. Meticulous scans the **actual rendered DOM** as it replays real user flows, so it catches problems that only exist at runtime: dynamically rendered content, state that appears after interaction (modals, menus, expanded panels), and issues introduced by third-party components or injected markup. Because it works from real sessions rather than source code, it surfaces regressions in the states your users actually reach — a layer of coverage a source-only tool cannot provide.