Multi-engine consensus
Five engines run in parallel, then a live-DOM re-verifier. Findings only count when they survive cross-engine de-duplication and re-verification.
axe · pa11y · ARC · IBM-EA · LH + Live-DOM verify
Forensic accessibility · WCAG 2.2 AA
Evidence-grade accessibility scanning for code that ships.
Five engines plus an independent live-DOM re-verification. Every patch is sandbox re-scanned against the live DOM before we call it VERIFIED. No overlays. No 200-page PDFs. No “full compliance” claims.
Engines wired in
axe-core IBM Equal Access ARC pa11y Lighthouse Live-DOM verifier
Exhibit A · Sample scan
Sample output
▍ ap-scan · https://your-site.com
● live
Score
87/100
Verified fixes
3 copy-as-PR
Tier-1 risk
1
Engines per scan
5
Consensus across axe, pa11y, ARC, IBM-EA, Lighthouse. Live-DOM re-verifies before patches ship.
WCAG 2.2 coverage
30 to 40%
Automated, honestly. The rest needs human review. We say so.
Sandbox revalidation
0 regressions
Patches only ship after a clean DOM re-scan. No competitor does this.
Section 02 · September 2026 field audit III
For the September 2026 field audit III we scanned 56 real sites twice. Then we checked our own homework.
Through production, 5 pages each, before and after a round of accuracy fixes. Then every finding the fixes removed was re-verified on the live page. Fixtures tell you what a scanner can do; the real web tells you what it does.
Findings, same 50 sites
928 from 926
0%, and no engine's count rose. Fewer rows because the same defect stopped being reported up to seven times.
Duplicate rows
0 from 0
One defect, one row, every engine that saw it listed as a voter.
Removed findings re-verified live
1,176
Each one hit-tested on the live page. 42 confirmed gone for the right reason; the 13 that were not became a fix of their own.
Section 03 · Capabilities
Built like an instrument, not a marketing page.
Every feature on this list is real and shipping. The rest of the industry sells overlays and 200-row spreadsheets. We ship working diffs.
Sandbox-validated patches
LLM patches re-applied to a fresh DOM and re-scanned. Only zero-regression diffs are emitted as Copy-as-PR. Otherwise: snippet.
apply · re-scan · only-then-emit
Legal-risk scoring
Findings weighted by ADA Title-III tier and EAA jurisdiction. Tier-1 criteria are cited in roughly 99 % of lawsuits, so fix those first.
ADA · EAA · EN 301 549 report
Shadow DOM + APCA
Shadow-root pierced, state-permutation axe runs, APCA computed on real rendered pixels. The accuracy bar no other SMB tool clears.
state-aware · APCA · pixel-true
Section 04 · Mechanism
The verified-fix loop.
A patch isn’t a fix until the page re-renders without the violation. Our pipeline proves that before you ever see the diff.
See the GitHub Action- 01
Detect
Engine consensus identifies a verifiable violation with stable selector + WCAG mapping.
- 02
Propose
LLM generates a unified-diff patch grounded in the captured DOM + rule context.
- 03
Re-scan
Patch is applied in a sandboxed clone of the page and axe-core re-runs against the patched DOM.
- 04
Verdict
Original violation cleared AND no new violations introduced → emit. Otherwise: fall back to snippet + manual notes.
- 05
Stamp
verifiedSurface to user with chain-of-custody: fetch timestamp, ETag, engine vote, verifier outcome.
Section 05 · Disclosure
“We will never say full WCAG compliance.”
Automated scanning catches roughly a third of WCAG 2.2 criteria. The rest takes human judgement, and no scanner can honestly tell you otherwise. We tell you what we detected, what we fixed, and what still requires human review. The result is a report your auditor will respect, and a diff your reviewer will merge.
Never
overlays
Never
unverifiable claims
Never
200-row spreadsheets