Why Reverter
Every audit-bound enterprise enters compliance season with a violation backlog measured in thousands. Rule-based scanners produce diagnostics, not fixes; runtime overlays leave source non-compliant and trigger US regulatory action. Reverter targets that gap.
Pain → Solution → Savings (per medium customer, 50 devs)
Order-of-magnitude estimates from independent industry sources. Reverter's end-to-end accuracy on production customer code is pending Patent A PoC v1.0 — we do not claim "validated on N customer sites".
| # | Validated pain | Annual cost without Reverter | Estimated relief with Reverter |
|---|---|---|---|
| 1 | 50 devs × ~30 a11y fixes / quarter × ~25 min / fix | ~2,500 dev-hours/yr (~$300–500K) | Tier 1 codemods short-circuit 35–50%; Tier 2 cache hit 80–90% post-warmup (Patent A spec, §IC1) |
| 2 | In-house a11y specialist hire ($120–180K loaded; 5K IAAP-certified worldwide) | $160K loaded | Reverter covers high-frequency rules; specialist time redirected to qualitative review |
| 3 | EAA fine exposure (varies by EU member-state implementation) | up to €100–250K per state, per finding | De-risk via documented evidence trail (audit log, SARIF export, EAA / WCAG 2.2 AA mappings) |
| 4 | Runtime overlay license (e.g., accessiBe) | ~$3K/yr/site + brand reputational risk + FTC precedent ($1M, 2024) | $0 — Reverter writes source code (Patent A IC4) |
| 5 | ADA Title III litigation (4,605 lawsuits in 2024; 69% e-commerce) | risk-weighted $5–20K/yr | De-risk via documented evidence trail + source-code remediation |
Sources: WebAIM Million 2025; Seyfarth ADA Title III Tracker; WCAG 2.2 + EN 301 549; Patent A spec §Background; FTC press release re: accessiBe consent order (2024).
External evidence
- WebAIM Million 2025: 96.3% of top one million home pages have detectable WCAG errors; average page carries 51 errors and 8.6 distinct rule failures.
- EAA (European Accessibility Act): in force 28 June 2025; binds private-sector digital products and services across the EU 27 to WCAG 2.1 AA via EN 301 549.
- ADA Title II (large entities): digital-accessibility deadline April 2026; small entities April 2027.
- ADA Title III lawsuit volume (Seyfarth, 2024): 4,605 lawsuits filed; 69% targeted e-commerce.
- FTC accessiBe consent order (2024): $1M penalty for misleading "AI-only" remediation claims; runtime overlays are a documented regulatory hazard.
- IAAP-certified specialists worldwide: fewer than 5,000 — supply cannot meet enterprise demand through hiring.
The visibility gap
A violation report tells you what is broken. It does not tell you what to type into the source file. The standard options are:
- Hire specialists — supply-constrained, slow.
- Pay an agency retainer — $50–500K/yr.
- Buy a runtime overlay — non-compliant source, FTC and lawsuit precedent.
- Train every developer — ROI 6–12 months, high turnover.
- Let an LLM propose fixes the developer accepts in their IDE — this is what Reverter delivers.
Why generalist AI assistants fall short
GitHub Copilot, Cursor, Claude Code, Windsurf, and Devin can all generate code. None of them ship the combination Reverter does:
- Tiered cascade. Generalist agents send every
prompt to the same model. Reverter's Tier 1 deterministic
rule engine handles 35–50% of WCAG violations
(decorative
alt="",lang="en",for/idassociation,aria-required-attr, duplicate-id) without ever invoking the LLM. - Structural-signature caching. Reverter's cache
key is content-independent
(
(violation_type, element_pattern, ancestor_chain, ARIA_role)), so the same fix template applies across pages and across tenants without exposing content. - Adaptive context windows. Color-contrast gets the element + computed colors; heading-order gets the full document heading tree. Per-WCAG-SC sizing, not one fixed envelope.
- PII-safe preprocessing. Microsoft Presidio + spaCy NER swap PII for type-preserving placeholders before the LLM call; restored on output. Round-trip non-PII byte-diff is zero by construction.
The competitive risk is real: Anthropic, GitHub, and Cursor will eventually ship generic accessibility fixes. The Patent A moat (filed 2026-04-06, 58 claims, 8 IC) targets the specific architecture that makes accessibility remediation cost-efficient and privacy-safe at scale.
Why current accessibility tools don't solve this
| Tool / approach | Diagnoses | Auto-fixes | IDE-native | MCP-native | PII-safe |
|---|---|---|---|---|---|
| axe-core / Pa11y / Lighthouse | Yes | No | Partial | No | N/A |
| accessiBe / UserWay (overlays) | Partial | Runtime overlay | No | No | No |
| Siteimprove AI Remediate | Yes | Yes | No (web-only) | No | Not documented |
| Deque axe MCP Server (2025) | Yes | No (hints only) | Via IDE MCP | Yes | No |
| Evinced MCP tools (Sep 2025) | Yes | No (detect only) | Via IDE MCP | Yes | No |
| AccessGuru / FeedA11y (academic) | Yes | Yes | No | No | No |
| Reverter | Yes (via @ariada/core) | Yes (Patent A IC1+IC4) | VS Code | Yes (IC6) | Yes (IC8) |
Sources: Patent A spec §Background; competitive analysis at
research/competitive-analysis/SUMMARY.md; MCP server
registry at github.com/modelcontextprotocol/servers.
Honest framing. Reverter implements a strict subset of Patent A. End-to-end accuracy on production customer code is pending the Patent A PoC v1.0 pilot wave. Numbers above describe the addressable problem, not measured customer outcomes.