1. One address
No account. Recon starts on a single page. It does not claim to see your whole organisation in twenty seconds.
Accessibility testing model
Your site.
Fewer barriers.
One first trail.
Trop ™ — a guide dog for WCAG, not another checker.
Send a page URL. The team runs a WCAG pass, persona agents and a short PDF back to your inbox.
- WCAG 2.1 / 2.2
- AI agents and personas
- PDF mini-report
Trop Recon Free · 1 URL
Audit. Report. Openness. WCAG 2.1 / 2.2 Recon · Scan · Audit People decide what the model may miss
How it works
Paste a URL. Get a trail back.
The model does the first pass. The team decides what to trust before anyone calls it an audit.
2. Personas walk it
Rules plus agents: a blind user, a keyboard-only user, someone who needs contrast and zoom.
3. A report lands
A short PDF by email. Scan widens the scope. Audit adds expert review and language a board can act on.
What ships today
Scan, personas and a report leadership can read.
This is live scope, used in real page tests.
- Automated WCAG 2.1 / 2.2. Labels, contrast, alternatives, structure, keyboard.
- AI agents and personas. Blind, low vision, keyboard-only, and more.
- PDF mini-report. First barriers, as Recon.
- Full report. Priorities, fixes and board-level wording, as Scan or Audit.
- Guide-dog persona. A trail you can follow, not a raw fail/pass dump.
Personas
The dog leads. Agents mark where the path breaks.
Trop ™ is not a person. It has a guide persona and agents that use the page the way different people do.
Trop guide
Keeps recon in one story. Marks the trail instead of dumping a log.
Blind user
Screen reader, focus order, control names, image text, keyboard traps.
Low vision
Contrast, zoom, spacing, text locked inside pictures.
Keyboard only
Tab, Enter, Esc, visible focus, skip links.
Technology and RAE
In the works since 2017. Built by RAE.
Bids to PARP, KFS and MARR did not land. The platform shipped anyway. AccessibilityTeam.eu runs the product. RAE owns the engineering.
RAE also shipped Migomatic, a sign-language translator, in the Inkubator Wielkich Jutra social-innovation programme.
Engine layer
One trail instead of five CSVs. We ingest what the field already runs: axe-core, WAVE, Lighthouse, HTML_CodeSniffer, Pa11y, IBM Equal Access, Accessibility Insights, QualWeb (ACT Rules), ARC Toolkit, ANDI, AccessLint, Asqatasun. Wrappers such as Playwright-axe, Cypress-axe and Lighthouse CI sit on the same bus. What an engine already flags, nobody types again.
Scrapers
Chromium via Playwright for a real layout, CSS and the accessibility tree. Puppeteer where a lighter pass is enough. jsdom or Cheerio for static HTML. Each run stores a DOM snapshot, computed styles, the a11y tree, a HAR and a screenshot. On re-audit we diff those artefacts so a contrast fix does not quietly break a heading or a focus order.
Personas
Agents walk routes, not just rules. Blind user with a screen reader. Low vision at 200 to 400 % zoom. Colour vision deficiency. Keyboard only. Mouse only, no keyboard. Motor / switch input. Voice control. Deaf user who needs captions. Cognitive load: short time, plain language. Vestibular: reduced motion. Mobile TalkBack or VoiceOver.
Models per criterion
Contrast uses APCA plus WCAG 2 ratios on computed colours. 1.1.1 checks alt presence and whether the text matches the image. 1.3.1 reads the heading and landmark graph. 2.1.1 and 2.4.3 build a focus graph. 4.1.2 aligns name, role and value with the visible label. 3.3.x maps label, error and help on forms. 2.2.x flags timers. 3.1.1 checks page language. Each criterion has its own model, not one generic LLM over the whole page.
Knowledge base
Descriptions of WCAG criteria, typical faults and recommended fixes. The model trains on fragments of our own reports and audits, then fills gaps from Understanding WCAG, ACT Rules and public techniques. New findings go back into the base. Re-audits teach which remediations held and which drifted.
Orchestrator
The core that ties the five layers. It queues the URL, runs engines and scrapers in parallel, normalises hits into one barrier vocabulary, drops duplicates, scores confidence, assigns a persona and a criterion, then writes the report chapters. Partners see that graph through the API and the dashboard.
API and dashboard are for partners. A URL queue, re-audit diffs, one barrier vocabulary, webhooks into Jira, GitHub or a CRM. Built for agencies and product teams that test a portfolio, not one site a year.
About the project
Trop ™ is an accessibility testing model. It finds the first trail. Scan orders the barriers. Audit adds a human and a decision.