RGAA (Référentiel général d'amélioration de l'accessibilité) is France's digital accessibility framework: the French translation of WCAG requirements into 106 checkable criteria. It applies to public services and, since 2025, to many private companies. The same logic applies wherever you operate: WCAG 2.x directly, or its regional derivatives such as EN 301 549 (EU) and Section 508 (US). An accessibility audit captures compliance at one point in time; regression testing is what keeps it in place from one release to the next.
Most sites that lose their RGAA compliance didn't lose it at audit time: they lost it six weeks later, when a component was redesigned, a theme was updated, or a brand color changed. Nobody re-ran the 106 criteria. The pattern holds the same way under WCAG 2.x, EN 301 549, or Section 508. This article lists the criteria that regress most often in practice, separates what an automated tool can check from what stays a human judgment call, and shows how to fold accessibility into a regression suite without writing code.
The essentials on one page (definition, no-code automation, how to choose a tool): Automated regression testing, no code.
Is your last RGAA audit more than three months old, and have you shipped since? Delta-QA replays your flows on every release and flags contrast, text, focus, and hidden elements that changed, without code. Try Delta-QA for free →
Why RGAA compliance erodes
An RGAA audit is a snapshot: an expert walks through a sample of pages and scores each criterion. The result is true on the day of the audit. After that, the site keeps changing:
- a design system update changes the color of secondary buttons, contrast drops below 4.5:1;
- a contact form redesign replaces labels with placeholders: the label disappears;
- a new menu component hides the keyboard focus indicator;
- an optimization pass strips "unnecessary"
altattributes off images that carried meaning; - a dark theme ships without its contrasts ever being checked.
None of these releases is an accessibility project. Each one regresses a criterion. That's exactly the mechanism of a regression: something that worked stops working, without anyone deciding it should.
The RGAA criteria that regress most often
In practice, a small share of the 106 criteria accounts for most post-release regressions:
| RGAA theme | Typical criterion | What breaks it in practice | Detectable automatically? |
|---|---|---|---|
| Color (3.x) | Text/background contrast ≥ 4.5:1; information not conveyed by color alone | New palette, dark theme, a "disabled" state that's too light | Yes, comparing rendering and computed colors |
| Images (1.x) | Relevant text alternative | Component redesign, an image turned decorative or the reverse | Partially, a missing alt yes, its relevance no |
| Forms (11.x) | Visible, associated label; identifiable error messages | Placeholder instead of a label, lost error state | Yes, form state and error rendering |
| Navigation (12.x) | Visible focus; consistent tab order | CSS reset, third-party component, outline: none |
Yes, capturing focus state |
| Presentation (10.x) | Readable at 200% zoom; content visible with no loss at 320px | Modified CSS grid, truncated text, overflow | Yes, rendering at several widths and zoom levels |
| Structure (9.x) | Consistent heading hierarchy | A component that downgrades an h2 to a div |
Partially, the structural change may or may not be visible |
| Mandatory elements (8.x) | Page language, page title | Modified template | Yes, displayed values and attributes |
Two takeaways. First, most RGAA regressions are interface regressions: things visible on screen (contrast, focus, truncated text, a broken form error). Second, part of it stays a human judgment call (whether an alt is relevant, whether a label is clear): a tool doesn't replace the audit, it keeps the audit from going stale.
What automated regression testing checks, and what it doesn't
Automated regression testing compares your baseline pages against their new state on every release. Applied to accessibility:
It catches: a contrast that drops, a text size that shrinks, a focus indicator that disappears, an error message that stops showing, a hidden element, a page that no longer fits at 320px or 200% zoom, a displayed value that changes (language, title).
It doesn't catch: whether a text alternative is relevant, whether a label is understandable, whether the reading order makes sense. Those call for an audit and testing with real users.
It complements DOM analysis tools (axe-core, Pa11y, Wave): those read the structure and flag a broken rule; regression testing checks the actual rendering and flags a change against a validated baseline (including when the structure is correct but the display has regressed). More on this: Regression testing and WCAG accessibility compliance.
An RGAA audit that stays true after every design system update? Record your baseline flows in Delta-QA; every release gets compared, contrast, focus, and states included. Free plan: 100 checkpoints per month, no credit card. Try Delta-QA for free →
Setting up RGAA regression testing without code
- Start from the audit. The pages in the audited sample become your baseline pages. Add the critical flows: contact form, authentication, search, main conversion path.
- Record the states that matter. A form empty, in error, and validated; an open menu; an element under keyboard focus; your dark theme if you have one. With a no-code tool, each state is a flow recorded by browsing.
- Cover the widths and the zoom levels. 320px (criterion 10.11), 1024px, and 200% zoom (criterion 10.4): the same scenario, several configurations.
- Replay on every candidate release. In staging before production; in production at a regular interval, for changes shipped without a deployment (CMS, widgets, CDN).
- Qualify every difference. The side-by-side report shows what changed; the team decides: a regression to fix, or an intended change to accept as the new baseline.
- Keep the human audit. A full RGAA audit is still needed on a regular basis and after every major change; regression testing is what keeps nothing degrading silently in between.
At Delta-QA, the comparison runs on a deterministic AI (not an LLM): a proprietary algorithm, with no learned model, that flags only what a human eye would notice and gives the same verdict on every run. For public bodies and regulated sectors, the On-Premise edition keeps screenshots and reports inside your own network. Sector guide: Regression testing for government and public-sector websites.
FAQ
What is the RGAA?
The Référentiel général d'amélioration de l'accessibilité is France's digital accessibility framework. It translates WCAG requirements into 106 checkable criteria, applicable to public services and many private companies. The same logic applies internationally under WCAG 2.x and its regional derivatives, such as EN 301 549 in the EU and Section 508 in the US.
Can an automated tool guarantee RGAA compliance?
No. Some criteria call for a human judgment call (the relevance of a text alternative, the clarity of a label). An automated tool catches the measurable regressions (contrast, focus, text, states, rendering at 320px and 200% zoom) and keeps an audit from going stale between reviews.
What's the difference between an RGAA audit and regression testing?
The audit evaluates compliance at a given moment, across every criterion. Regression testing checks, on every release, that your baseline pages haven't changed compared to the audited state. One establishes compliance, the other maintains it.
Do you need to code to automate RGAA regression testing?
No. With a record-and-replay tool, flows and states (a form in error, keyboard focus, a dark theme) are recorded by browsing, then replayed and compared automatically.
Does regression testing replace axe-core or Pa11y?
No, it complements them. DOM analysis tools flag a rule broken in the structure; regression testing flags a change in the actual rendering compared to a validated baseline.
Your next release shouldn't regress your RGAA compliance. Create a free account, record the pages from your audit as your baseline, and let Delta-QA compare every release. Try Delta-QA for free →