Questions and glossary

Straight answers, plain words

The questions people ask before they trust a tool, and the words you will meet in an Axcess report.

FAQ

Common questions

Does Axcess certify that my website is accessible?

No. Axcess gives you evidence to review. It can't certify WCAG conformance or prove legal compliance, and it doesn't replace testing with people who use assistive technology. An accessibility specialist can use its evidence when writing a conformance report.

What do Barrier, Needs review, and Informational mean?

They are the three report groups, and every result lands in one of them. The glossary below defines Barrier, Needs review, and Informational, and What Axcess checks shows which checks feed each group.

How is Axcess different from Siteimprove or axe DevTools?

We use and like both. Siteimprove monitors whole sites from the cloud, and the axe DevTools browser extension checks the page in front of you. Axcess runs on your computer, scans behind a sign-in including two-factor steps, opens menus and dialogs by itself, and adds browser checks such as reflow and text spacing. See the side-by-side comparison.

Do I need AI to use it?

No. The rule engines and the keyboard, zoom, focus, and click-through checks need only a browser, which the desktop app includes. AI-assisted checks are optional, run on a local model you install yourself, and are never reported as Barriers.

Does any of my data go to the cloud?

No. There is no account, no telemetry, and no upload. Axcess connects to the website you are scanning and, if you install one, a local AI service on the same machine. The desktop app also checks GitHub once per launch for a newer version, and that request carries no scan data. Read the privacy page.

Can it scan pages behind a login or two-factor sign-in?

Yes. You sign in directly with the website in a browser window that Axcess opens, and the scan continues with that session. Axcess never sees your password or code. The site must use HTTPS and a public address, and you need permission to test it.

What does "29 of 55 criteria" mean?

Axcess has at least one check that gives evidence for 29 of the 55 WCAG 2.2 Level A and AA success criteria. That does not mean those 29 are decided without a person, and it does not mean a site that passes them conforms. The remaining 26 are listed as manual, with the steps to test each one. See all 55 success criteria.

How accurate is it?

Every result carries its method and report group, so you can see how certain it is, and you can mark any result as a false positive with a reason. Only rule-engine failures are reported as Barriers; everything else waits for a person. The project also checks its results against a set of made-up examples, as a safety rail rather than a measure of accuracy on real sites. How we measure accuracy.

How long does a scan take?

It depends on how many pages you scan and which checks you turn on. Start with about 25 pages and grow from there. Speed and coverage tips.

Will it break anything on the site?

Axcess reads pages and, unless you turn it off, operates visible controls such as menus and tabs. While it clicks, the click-through:

never follows links to other addresses
refuses controls named sign out, delete, remove, unsubscribe, submit, save, and similar words
blocks form submissions and other changes
undoes navigations
stops after a fixed number of clicks per page

It is a safety net, not a guarantee, so use a test or staging copy of a site when you can. Public scans respect robots.txt by default.

Does it work on single-page apps built with React, Vue, or Angular?

Yes. By default, Axcess renders each page in a real browser before testing, follows app-style routes as separate pages, and can open menus and dialogs that only appear after a click. Routes must be reachable by real links inside your scope; Axcess does not read application code to guess private addresses.

Can several people use one Axcess?

A small trusted team can share one instance on an always-on machine over a private network with a shared access token. It runs one scan at a time and must not be exposed to the public internet. Larger, stricter institutional deployments are described in the documentation.

What does it cost, and who maintains it?

Axcess is free and open source under the MIT license. It was built at the University of Michigan's College of Literature, Science, and the Arts. The source, documentation, and white paper are on GitHub.

Which platforms are supported?

The desktop preview is available for Apple Silicon Macs and 64-bit Windows. From source, Axcess runs on macOS, Linux, and Windows with WSL. A Linux desktop installer is part of the build system but is not yet released as a preview.

Glossary

Words you'll see in a report

These definitions come from the Axcess glossary in the project documentation, so the site and the documentation always say the same thing.

Report groups

Barrier: A result where a rule engine (axe-core or Siteimprove Alfa) failed a fixed, machine-testable rule, such as an image with no alt text. These rules give the same result every time, so these are the most certain results. Start here: check that the rule applies to that part of the page, fix it, then rescan to confirm the fix.
Good to know: some Barriers are best practices, not WCAG failures. The Issues table's WCAG column shows Best practice for these.
Needs review: A possible problem, found by a less certain check, that a person must confirm before it counts as a Barrier. Open the page and check the item yourself. If it is a real problem, record that in its status and fix it; if not, mark it as a false positive with a short note.
Found by: a browser check (such as focus, zoom, or reflow), the keyboard trap check, a motion check, image text whose alt text is missing or does not match, a local AI model (such as for link text, headings, or form labels), or a Siteimprove Alfa "cannot tell" result.
Informational: A record kept so you can see what was checked, not a problem to fix. You do not need to act on it, and it should not be reported as an issue.
Examples: text in an image whose alt text already says the same words, and results from older checks that are no longer considered reliable.

Issues and findings

Finding: One result a check recorded, with its page, element, and evidence. The app calls each one an occurrence; exports and developer tools still say finding.
Occurrence: One place a problem appears: one element on one page, or in one page state. The Issues table counts these in its Occurrences column, and the Excel workbook calls them Instances.
Issue: One row in the Issues table: every occurrence found by the same check (for images, the same kind of image with the same alt text problem). The Issues page shows the number of issues and occurrences side by side.
Found by: Which kind of check found an issue, shown in the Issues table's Found by column and filter.
WCAG: found at page load by a rule engine (axe-core or Siteimprove Alfa), a browser check, or a local AI model.
Click-Through: found only in a page state opened by clicking a control, such as a menu. It shows only when at least one of the issue's occurrences needed the click. One issue can be both WCAG and Click-Through when the same problem appears at page load and behind a control.
Alt Text: text found in an image, and whether its alt text says the same thing.
Root cause: The underlying reason a problem repeats, such as one template or shared style file (stylesheet) used on 40 pages. Axcess groups occurrences by the check that found them, not by root cause, so look for a shared source before fixing page by page.
Evidence: What a check actually recorded: the page, the element, a code snippet, often a screenshot, the rule, and the method. It stays with the report so anyone can check a result later.
Element: One piece of a web page, such as a heading, image, link, button, or form field.

Severity and priority

Impact: How badly a problem affects people: Critical, Serious, Moderate, or Minor (for local AI checks, this rating shows how confident the model is). Image checks say major for Serious and info for Minor, and the workbook and written report show Siteimprove Alfa results, which have no rating, as Moderate.
Priority: A High, Medium, or Low label in the Issues table that combines impact with how many pages an issue touches. Because it rewards spread, a severe problem on a single page shows as Low, so check its impact too.
Status: Where an occurrence stands in review: New, Reviewing, In progress, Fixed, Accepted risk (a known problem your team chose not to fix for now), or Not a problem (a false positive). The last four need a short reason. The Excel workbook uses Not Started for New, Resolved for Fixed, and Not an Issue for Not a problem.

WCAG terms

WCAG: The Web Content Accessibility Guidelines, the international standard for accessible web content. Axcess checks against WCAG 2.2.
Success criterion: One testable requirement in WCAG, numbered like 1.4.3 (Contrast Minimum). WCAG 2.2 has 55 success criteria at Levels A and AA.
Alt text: The written description of an image that screen readers read aloud, and that appears if the image does not load. WCAG calls it a text alternative.
Conformance: A page conforms to WCAG when it meets every success criterion at a chosen conformance level. A scan cannot prove conformance, because many success criteria need manual testing.
Conformance level: WCAG sorts success criteria into Level A (the minimum), AA (what most policies require), and AAA (the strictest). Axcess checks Level AA by default, and you can choose A or AAA when you start a scan.
Best practice: A result that is good practice but not tied to a WCAG success criterion, labeled Best practice. When a rule engine finds one, Axcess lists it with Barriers, so fix WCAG Barriers first and treat Best practice items as recommended.

Coverage

Scan coverage: What a scan actually checked: which pages it tested, which checks ran, and how many page states it reached. The report shows this under What was checked, above the Issues table.
Crawl: How a scan moves through a site: Axcess starts at the address you give, follows links to find more pages, and tests each page in the scope.
Pages not reached: Pages the scan tried but could not load, which the report summary counts as errors next to the number of pages checked; a page that answered with an error, such as "Sign-in required", is listed with that status instead. The report does not list pages skipped because the site asked scanners to stay out (its robots.txt file) or because they were outside the scope.
Scope: The part of a site a scan may visit: pages under the address you start from, such as everything under www.example.edu/admissions/. A page limit and a limit on how many links deep the crawl goes also apply.
Page state: How a page looks at one moment: "At page load", or after a control is used, for example after a menu opens. Problems that appear only after a click are labeled "After clicking" with the control's name, so you can reproduce them. Developers call this a DOM state (the DOM is the browser's live copy of the page).
Click-Through: The check that opens menus, tabs, dialogs, and other controls, then checks the page again in each new page state it reaches. It is on by default. On the New scan form it is Open menus, tabs, and pop-up windows (Click-Through).

Accuracy

False positive: A result that turns out not to be a real problem. Mark it as a false positive with a short reason, and Axcess keeps that decision with the report and moves it out of the written report's worklist.
Zero false positive goal: Our goal is that nothing Axcess reports as a Barrier is a false positive, which is why only rule-engine failures go there and every other check waits for a person. It is a goal, not a guarantee, so confirm each Barrier on the page before you report it.

Testing methods

Automated testing: Checks a program runs on its own, such as a rule engine or a browser measurement. They are fast and consistent, but they only test what a machine can measure.
Manual testing: A person checks the site directly, for example using only a keyboard or listening with a screen reader. Many WCAG success criteria can only be judged this way.
Screen reader: Software that reads a page aloud, or shows it on a braille display, for people who are blind or have low vision. JAWS, NVDA, and VoiceOver are common screen readers.
Keyboard focus: The item a keyboard user is on right now, usually shown with an outline. Pressing Tab moves focus to the next link, button, or form field.
Axcess and manual testing: Axcess speeds up manual testing by finding the machine-testable problems and pointing you to the pages, elements, and states that need a closer look. It does not replace manual testing, because many success criteria have no automated check at all; What Axcess checks lists them.

Checks and tools

Rule engine: Software that tests a page against a fixed list of machine-testable rules. Axcess runs axe-core by default and can also run, or instead run, Siteimprove Alfa.
axe-core: The open-source rule engine from Deque Systems. By default, Axcess runs a bundled copy on every rendered page.
Siteimprove Alfa: Siteimprove's open-source rule engine, built on ACT rules, which Axcess can run on your computer alongside or instead of axe-core. When Alfa cannot decide a result by itself, it reports "cannot tell", and Axcess puts that result in Needs review.
ACT rule: An accessibility test written in the Accessibility Conformance Testing format from the W3C (the group that publishes WCAG), precise enough that different tools can run it the same way.
Browser check: A check that measures how a page behaves in a real browser, such as resizing it to phone width or pressing Tab through it. Its results go to Needs review.
Motion check: A check for audio that plays by itself with no control, and for autoplaying video or scrolling marquee text with no way to pause it (WCAG 1.4.2 and 2.2.2). Its results go to Needs review, and in the app it runs only when you turn on Check motion and animation.
Local AI model: An optional AI model that reads text (a language model) or looks at images (a vision model), running on your own computer through a free program called Ollama. Axcess never installs one for you, and AI results are never reported as Barriers.
OCR: Optical character recognition: software that reads text inside images. The desktop app includes the Tesseract OCR engine; if you run Axcess from source code, install Tesseract separately.
Rendered page: A page after the browser has run its scripts and drawn it, which is what visitors see. Axcess tests rendered pages by default.

Common problems

Image of text: Text that is part of a picture instead of real text (WCAG 1.4.5). Screen readers cannot read it, and people cannot resize or restyle it.
Keyboard trap: A spot where keyboard focus gets stuck and Tab or Shift+Tab cannot move it away (WCAG 2.1.2). People who do not use a mouse are stranded there.
Reflow: Content should fit a window 320 CSS pixels wide (about a small phone, or a desktop browser zoomed to 400%) without scrolling sideways (WCAG 1.4.10). A CSS pixel is the browser's unit of measure, not a physical dot on your screen.
Resize text: Text should stay readable, with nothing cut off, when zoomed to 200% (WCAG 1.4.4).
Text spacing: Text should not be cut off when a reader increases line, letter, word, and paragraph spacing (WCAG 1.4.12).
Focus not obscured: A control that has keyboard focus should not be completely hidden behind something else on the page, such as a sticky header that stays on screen while you scroll (WCAG 2.4.11).
Target size: Buttons and links should be at least 24 by 24 CSS pixels, or have enough space around them (WCAG 2.5.8, Level AA). The stricter 44 by 44 CSS pixel size is WCAG 2.5.5, Level AAA, which Axcess checks only when Siteimprove Alfa runs at Level AAA.

Using Axcess

Sign-in scan: A scan of pages behind a sign-in. Axcess opens a browser window, you sign in yourself (including any two-factor step), and Axcess then scans the site as you, signed in, without ever seeing your password.
Rescan comparison: Two reports of the same scope lined up on a report's Compare reports view, with each issue marked New, No longer found, or Still found. No longer found means only that the later scan did not find the issue again. It is not proof of a fix, so check the page yourself before you mark the issue Fixed.
Draft export: An export downloaded from the app before expert review is finished. The app marks it DRAFT in the file name and inside the file; command-line exports are not marked.
Configured search: An optional scan setting, Use a search box to find more pages, that types a sample search you choose and tests the result pages it finds. Single-page apps and search scans explains how to set it up.
Written report: The report you download as Written report (Markdown) from the Export report menu. It is a plain text file with headings (Markdown format). The Excel workbook is a separate export, Issue list with fixes (Excel), with one row per issue.
Local-first: Axcess stores your reports on your computer, has no account, and sends no usage data back to us (no telemetry). It connects to the site you scan, and the desktop app checks GitHub for updates.