How the review is done, in full

Scope

One domain. Five public URLs, agreed with you in advance — typically the home page, a listing page, a detail page, one form or checkout step, and contact. Public pages only: we never ask for and never accept credentials, and no authenticated flow is tested.

Pass one — automated

Static analysis of each page's markup for checks that can be decided from the markup alone:

• Images without an alt attribute
• Form controls with no label, aria-label, aria-labelledby or title
• Missing lang on the document
• Heading levels that skip
• Missing or empty page title
• Duplicate element identifiers
• Links and buttons with no discernible text
• No skip-to-content link
• Positive tabindex overriding natural focus order
• No main element or role="main"
• Link text that does not describe its destination
• Image buttons with no accessible name

Pass two — browser verification

Each page is loaded in a real browser and tested for the things markup cannot tell you:

• Computed colour contrast against the actual rendered background, at the actual font size and weight
• Keyboard traversal: tabbing the full focus ring and detecting genuine traps
• Focus visibility, and focus order against DOM order
• Reflow at a 320-pixel viewport
• Text resized to 200%
• Form error identification: whether validation errors can be announced at all
• Target size for interactive controls

The quality rule

Every automated finding is reproduced in the browser before it appears in your report. Anything that cannot be reproduced is discarded and is not reported to you.

Findings repeated across pages by a shared template are grouped into one issue listing the affected pages, rather than counted five times. You are told how many distinct defects you have, not how many times we saw them.

Declared but not performed

We do not perform a screen-reader spot check. It is named here rather than implied, because a method that quietly omits something it sounds like it covers is worse than one that admits the gap.

Limitations, stated plainly

• Only the listed pages are tested, not the whole site
• One browser engine and viewport set; other browsers and assistive technologies may differ
• Public pages only. No authenticated or account-holder flows
• A point-in-time assessment; it says nothing about your site tomorrow
• Automated and scripted testing detects only a minority of accessibility barriers
• Not a legal opinion and not a conformance certification

Quality assurance

Reports are produced by an AI agent and quality-assured by Antony Draper on a sampled basis: the first three reports in full, then one in five, plus every report that draws a refund request, a complaint or a dispute, and any report the agent flags as uncertain.

If you want the work done instead

You can fix these yourself, use your own developers, or any supplier you choose. If you would like us to quote for the remediation, ask and we will. We will not follow up to offer it.

Disclosure

Testing and report drafting are performed by an AI agent (Claude), using automated tooling and a browser-driven verification pass. Antony Draper is the supplier of record and is accountable for the delivery, the quality and the accuracy of every report issued, and for refunds and complaints. Reports are quality-assured on a sampled basis before release. It is a technical assessment against WCAG 2.2 AA — not legal advice, not a certification, and not a guarantee of compliance. Whether the European Accessibility Act applies to you depends on your jurisdiction, your organisation's size and the services you provide.