This site is scanned by the engine it sells

Scope. Every page on klarvoaccess.com. Not a sample, not the home page — every route in the sitemaps, which is how the gate below discovers what to scan.

Standard. WCAG 2.2 Level AA as the floor, with Level AAA where it costs nothing to be better. In practice that means body text at a contrast ratio of at least 7:1 rather than the 4.5:1 the standard asks for, a visible focus indicator on every interactive element, primary controls at least 44 by 44 pixels, full support for reduced-motion preferences, and a layout that reflows at 400 per cent zoom without horizontal scrolling.

Technique. No accessibility widget, no overlay, no consent banner. The site is static HTML with one small piece of JavaScript — the scan form — which works as a plain form if that JavaScript never loads. There are no cookies and no third-party requests of any kind, which is enforced by a build gate rather than by good intentions.

Date. This statement describes the site as deployed. It is not a certificate: our own automation cannot issue one, and neither can anyone else's. Conformance for a scope and a date is established by a human audit, which is a service we sell and which we are not going to pretend a build script replaces.

A statement with no known issues in it is a statement nobody checked. Ours is a short story with a good ending, and the whole of it stays on this page:

An earlier run of our own product against this site returned thirty-one findings on the home page. We read every one, concluded that thirty of them were defects in our scanner rather than in the site, and published that conclusion here. Then we fixed the scanner:

• 2.5.8 Target Size had flagged twenty-eight ordinary links inside sentences. WCAG exempts targets in a block of text; the probe now applies that Inline exception before counting, and names it in the finding it emits. We kept the change the false findings prompted anyway — inline links here carry vertical padding, so their pointer target is at least 24 pixels tall even though the standard would let us leave them at sixteen.
• 1.4.12 Text Spacing had counted containers that were already clipping by design. The criterion is about content lost because the spacing adaptation was applied, so the probe now measures before and after and counts only new clipping.
• The same pass caught two wrong levels. The engine carried 3.2.6 Consistent Help and 3.3.7 Redundant Entry at Level AA, from earlier drafts of WCAG 2.2. The published Recommendation makes both Level A, and every finding and conformance table now says so.
• And it surfaced a defect the findings could not name: this site's strict Content-Security-Policy was silently blocking the three injected engines, so scans of this site ran on the probes alone and reported themselves failed. The scanner now bypasses the target page's CSP for its own instruments — what any auditor's tooling does, changing nothing about what a visitor receives — which matters for every hardened store, not just ours.

The most recent run of the fixed engine against this site: six pages, all three engines and the full probe set on every one, zero violations, and exactly one finding — 3.2.6 Consistent Help (Level A), a heuristic noticing that our contact route appears in the footer everywhere and also in the body of some pages. Its own description asks a human to confirm the placement is intended. We have, and we are content with it.

We could have quietly published none of that. Publishing it is the more useful thing, and finding four product defects by pointing the product at ourselves is the only version of the dogfood claim that is actually true.

The rest of what we know:

• Automated coverage is partial, here as everywhere. Of the 55 WCAG 2.2 A and AA success criteria, 39 have at least one automated rule or probe behind them and 16 have none. Passing our gate means no machine-detectable violation, which is not the same as conformance.
• The per-route product scan is not yet automated end to end. The public scan endpoint applies a ten-minute cooldown per target host — correctly, so the funnel cannot be used to hammer anyone — which means it cannot scan twenty-five routes of one site in a sitting. Running the full set needs the authenticated scanning path a paying customer's monitoring uses, and that is the next piece of work on this page. Until then the per-route gate is the rule-set walk above, and the product scan is run on a curated set.
• This site has not had an independent third-party audit. We audit customers' stores; nobody outside has audited ours. Saying so seems more useful than the alternative, and it is on the list.
• The hero animation is decorative. It plays once and is fully disabled when your system asks for reduced motion. If it ever fails to respect that preference, that is a defect — tell us.
• Long tables scroll horizontally inside their own container on narrow screens. The page never scrolls sideways, but a wide comparison table does, which is a compromise rather than an ideal.