Honest coverage
Axcess contributes evidence to 29 of 55 WCAG 2.2 A/AA criteria, and says so. Every public number comes from the same file the product reads, so a claim can never drift from the code.
About Axcess
Evidence before verdicts
Axcess started with one hard problem and grew into a way of working. This is the story, the idea that guides it, and where it is heading.
The beginning
One problem nobody's scanner could solve
An accessibility lead at the University of Michigan needed to find every image that was really just a picture of text, across an entire website, before the next reporting cycle. Text inside an image cannot be resized, restyled, or read aloud, and existing tools could not find it reliably at scale. Checking every page and image by hand was slow, repetitive, and easy to get wrong.
The first version of Axcess did exactly that job. It crawled a site without sending anything to the cloud, found the images, read the text inside them, compared it with the alternative text on the page, and ranked the likely problems for a person to review. Then it could scan again to confirm the fixes.
The intent was never to remove the expert. It was to take away the repetitive discovery work and hand the expert better evidence.
How it grew
Finding a problem is only the first question
A raw result does not answer what people actually ask. What happened? Who is affected? Which pages? How certain is it? What should change? How will we know it is fixed?
So Axcess became an evidence workbench. New checks were added for page structure, names and roles, keyboard behaviour, focus, small screens, media, and meaning. A manual evaluation workflow was added because many accessibility questions cannot honestly be answered by software. And the product became explicit about uncertainty: a rule failure, an observed behaviour, and an AI suggestion are never treated as equally certain, and each keeps the name of the method that produced it.
The guiding idea is simple: preserve the evidence before presenting a verdict.
Principles
What the project holds itself to
Private by default
Evidence stays on the auditor's computer. No telemetry, no cloud AI, no account. Sensitive and login-protected pages deserve nothing less.
Universal design first
The tool's own interface is held to WCAG 2.2 AAA and designed around an accessibility professional who may use a screen reader, magnification, or keyboard only. When efficiency and universal design pull apart, universal design wins.
Human decisions, recorded
Automated results are input, not output. People confirm, reject, and remediate, with reasons and history, and the report says what was and was not evaluated.
Direction
Where Axcess is heading
The project's white paper lays out its long-term direction: an offline, evidence-first accessibility quality system, grounded in the W3C's evaluation methodology and ACT rule format.
Truthful, inspectable coverage
Record exactly what happened for every page, state, and method. "Found nothing", "skipped", and "unavailable" must never look the same.
Test experiences, not just URLs
Safe, user-authorised journey recipes so the states after a click, a form error, or a sign-in are tested too.
Issues as durable work
Separate what was observed, what people decided, what was remediated, and what a comparison showed.
The report as the product
A self-contained, accessible HTML report that works without Axcess running and answers what was evaluated, what was not, and what to fix first.
"Fully offline", defined
A private live scan that touches only the target, and an air-gapped replay mode that touches nothing.
Validation before promotion
Every new detector measured against expert-reviewed real-world examples before it is trusted.
Open source
Built in the open at the University of Michigan
Axcess is MIT licensed and developed by the LSA Technology Services team at the University of Michigan's College of Literature, Science, and the Arts. The crawler, evidence store, interface, and exports are all public.