We publish the proof, not just the pitch.

We test the scanner on fresh stores and have the results audited independently. The latest frozen test is below our release bar, so the work continues. Here is what it already covers, how it performed, and the bar it has to clear.

Website scans completed

The live total counts every completed scan. Store websites are shown separately, so repeat checks never masquerade as broader coverage. No merchant identity is published.

Total website scans completed: —
Store websites scanned: —
Totals checked: —

Loading. Loading completed website-scan totals.

Performance on fresh test stores

Measured 2026-09-22 on fresh stores none of us had scanned before, and generated from that run’s hash-verified result.

Below release bar — work continues

Observations judged correct: 99.33%
Unclear observations held back: 18.03%

What this number is

Every observation the scan published on those stores, checked one by one against evidence captured before the scan ran. It is a count of individual observations, not a promise about any single report.

Why we hold evidence back

Bot blocks, missing pages, ambiguous wording, and incomplete captures do not support a fair yes or no. Those calls stay unknown instead of becoming a confident-sounding guess.

Tested across markets and languages

We test unfamiliar storefronts across markets and languages, so the scanner has to earn its keep beyond one familiar shop. The independently audited places and site languages below are the coverage we can prove.

Countries represented in the latest test

Site languages represented

Where the scanner was tested is one question. Which languages its own rules can read is another, and the honest answer is check by check. Each check carries its own word list, and all 3 of them read 36 of the 37 languages in the table, covering 69 markets. Everywhere else at least one check reports that it could not look. See the check-by-check table.

Public benchmark projection: frozen 2026-09-22, result SHA-256 91fbac988008a9d7358889756cdbe52ae097fbd9cd049a468fd6852e557c3231. The benchmark is a dated test result, not live scanner performance.

What we could not check

A check we could not decide is published as its own row and counted apart. It scores nothing either way — our gap must never cost a store points, and it must never be mistaken for a clean bill of health. These are the reasons a report can give, and it gives one of them every time.

The page was not found: We looked for the page this belongs on and did not find one, so there was nothing to read.
The page does not say: We read the page and it does not state this.
The page is unclear: The page says something about this and we could not settle what it means, so we did not guess.
We read only part of the page: We captured only part of this document, and a fragment is not enough to judge it on.
Nothing linked to the page: Nothing on the storefront linked to this page, so we could not follow the route a shopper would take.
Nothing to open: Neither the sitemap nor the navigation gave us an address to open for this.
Nothing we opened was the right page: We opened what we found and none of it turned out to be the page this check needs.
The page never answered: We asked the store for this page and it did not answer.
The store turned us away: The store turned us away before we could look. That is a wall in front of the page, not a missing page — and Google’s crawler can meet the same one.
The checkout was never reached: We never got into the checkout, so nothing that can only be seen there was judged.
The market is unknown: We could not work out which market this store sells to, and this check needs that before it can judge a national format.
We do not read this language yet: This check finds its subject by matching words, and it does not yet have the words for this page’s language. That is our gap, not yours.
Nothing outside the product name backs it up: A word on the page pointed at this, but it sat in a product name or an option label. Your own categories, menus and catalogue say nothing of the kind, so we did not treat it as a finding.
Part of the scan did not run: The part of the scan that would have decided this did not run.
We could not look: This scan could not look. Nothing was found, and nothing was ruled out.

See how the scanner gets better

Every published update ties a real improvement to the report a merchant actually sees.

See the latest improvements