Methodology
Three layers, one signature
We publish how the audit works because the reader we optimize for is skeptical by profession: your defense attorney. Every finding in the report says which layer produced it.
Layer 1: deterministic machine checks
Every audited page is loaded in a real browser. We run the industry-standard axe-core rule set (the same engine professional auditors run), plus checks of our own that the standard tooling omits: live keyboard probing (tab order, focus visibility, focus traps), DOM integrity checks (broken skip links and in-page anchors), and a full evidence capture: screenshots, element inventories, computed accessible names.
Checks the machine layer cannot decide conclusively, for example text contrast over background images, are not silently dropped and not silently guessed. They are flagged for human review and resolved by the reviewer, and the report shows them as their own section.
Layer 2: criterion-scoped AI judgment
Roughly a third of WCAG is machine-testable. The rest requires judgment: is this alt text meaningful, does this link text convey its purpose, is this heading structure honest, is the same control named consistently across pages. Our AI layer evaluates exactly these criteria, one criterion group at a time, always against captured page evidence, never from memory. Findings state facts and cite the offending element; uncertainty is stated in the evidence rather than hidden.
Layer 3: certified human review and named sign-off
An IAAP-certified accessibility specialist reviews every report before delivery: a false-positive pass over every serious finding, resolution of all items flagged for human review, targeted keyboard and screen reader spot-checks aimed at the criteria where automation is weakest, and a severity sanity check. The reviewer can strike any finding and reject any report. The report ships with the reviewer's name and credential on it.
The benchmark
Before selling anything, we ran the full pipeline against reference sites whose accessibility failures are publicly documented by professionals, including the W3C's deliberately inaccessible demonstration site with its complete expert evaluation reports, and measured our detection rate criterion by criterion. We iterated until the pipeline met our bar and we keep re-running the benchmark as the pipeline evolves. Where automation is structurally weak, those exact criteria are targeted by the human reviewer's protocol instead of being ignored.
What the report claims, and what it does not
The report is an independent technical assessment against WCAG 2.1 AA. It never claims your site is "compliant" or "certified", because no honest audit can. It gives you per-page findings with criterion mapping, severity, evidence and remediation steps, and it states transparently what was tested by machines, what was judged by AI, and what the human reviewer verified.