What we check
Eight areas, measured and sealed separately
A website can pass accessibility and fail on loading performance. Reporting that as one combined figure would hide exactly the information you need, so we never produce one. Each area below is measured on its own, judged on its own and — if it reaches 90 out of 100 points and control over the domain has been proven — sealed on its own.
The eight areas
Accessibility (WCAG 2.2 AA)
55 success criteria
Measured in a real browser across three device profiles. A substantial share of these criteria cannot be decided by a machine — those stay visibly open until a human auditor has ruled on them, rather than being counted as passed.
Search engine optimisation
15 rules
Titles and descriptions, heading structure, canonical addresses, structured data, social previews, image alternatives, internal linking and language markup.
Visibility in AI answers
9 rules
Whether an answering system can find, extract and attribute a passage of your page: answer-first openings, question headings, self-contained sections, author and freshness signals, machine-readable markup.
Loading performance
10 measured signals
Seven metrics (LCP, FCP, INP, TBT, CLS, TTFB, Speed Index) plus the separation of field data from real visitors and lab data under fixed conditions, and mobile separately from desktop.
Data protection & cookies
10 weighted rules
What loads before any consent, whether pressing “reject” actually stops it, cookies set before a choice was made, remotely loaded web fonts, and the presence of a privacy notice. This is the one place where the auditor does more than read — it clicks “reject” and reloads, because that question cannot be answered by reading.
Indexing & technical setup
9 weighted rules
Whether search engines can reach, read and index the page: status codes, noindex directives, redirect chains, canonical addresses, content present in the delivered HTML, and whether robots and humans receive the same page.
Responsive behaviour
8 measured signals
Three device profiles rendered in a real browser — not a narrowed desktop window. Horizontal overflow, target sizes, text overflow, content visible without scrolling. Every finding is measured twice; only what both runs see is reported.
Software currency
4 weighted rules
What the site reveals about its own software — front-end libraries past their maintenance end, version numbers exposed in markup or headers. This is not a security scan: it never probes paths, never attempts a login, and issues no severity above “serious”.
Rules or signals — why the difference matters
Six areas rest on a fixed, countable set of rules: each rule has an identifier, a weight, and appears by name in the report, so a finding can be traced back to the rule that produced it. Two areas rest on measured signals instead — values read from a real browser, where there is no finite rule list to enumerate. Both are honest; they are simply different kinds of measurement, and the report says which is which.
In the three areas with weighted rules, the weights add up to exactly 100 points, so the result reads as a percentage. The weighting follows actual risk rather than ease of measurement — in software currency, for example, one rule carries 40 points and the other three 20 each, because only the first describes a real defect.
Where a machine stops
Accessibility is the area where automated testing reaches its limit. Whether an alternative text actually describes the image, whether an error message is understandable, whether the reading order makes sense — a human decides that. Tools that still certify “100 % accessible” after a clean automated run are hiding that gap. Here such criteria stay visibly open, and accessibility can only be sealed at the level that includes human judgement.
The opposite pole is indexing: every criterion there is unambiguously machine-decidable. A status code is a status code.
The full list of every individual check point is published — in German, one page per area: Leistungen →