#accessibility
19 posts
Consistent Help, Computed: What an AI Design Agent Can and Cannot Check in WCAG 2.2 SC 3.2.6
SC 3.2.6 Consistent Help is Level A, and its documented test runs on serialized order. A computable check for AI design agents, plus its limits.
Error Messages as a Computable Contract: Text, Bound, Flagged, Corrective
WCAG 3.3.1 and 3.3.3 plus MDN's aria-invalid and aria-describedby split a form error into four clauses. Three can be checked in the DOM; the fourth needs judgment.
The Contrast Rule an Agent Can Actually Run (and the One It Can't)
WCAG 2.2 SC 1.4.3 and APCA 0.0.98G-4g are both computable from CSS values, but the WCAG 3.0 Working Draft still leaves its contrast measure undetermined — so an honest agent refuses that third verdict.
WCAG 3.3.8 Accessible Authentication (Minimum): A Computable Checklist for Login Flows
Success Criterion 3.3.8 restated as seven machine-checkable signals an AI design agent can flag in a login flow, with the boundary of what only human verification can settle.
Accessibility Statements Under the EAA: What an AI Design Agent Can Verify and What It Cannot
An EAA accessibility statement is a set of claims an agent can partly check — structure, drift, version, dates — and a conformance assertion it must refuse to certify.
WCAG 1.4.10 Reflow: A Computable Check for Design Agents
How AI design agents can compute WCAG 1.4.10 Reflow: the 320 CSS px test, viewport meta rules, ACT rule b4f0c3, and the exempt-table pattern.
WCAG 2.5.7 Dragging Movements: A Computable Contract for AI Design Agents
WCAG 2.5.7 Dragging Movements: what an AI design agent can detect mechanically, what needs a human rubric, and the single-pointer alternatives W3C documents.
Computing Translucency: Which Glass Decisions an Agent Can Read From Computed Styles
Liquid Glass and glassmorphism are usually judged by eye. Five checkable properties — layer, blur radius, tint alpha, worst-case contrast, and a reduced-transparency fallback — can be read from computed styles instead.
Accessibility Compliance for AI Design Agents: EAA, EN 301 549, WCAG 2.2 AA, and What Your Tools Can't See
What the European Accessibility Act, EN 301 549 and WCAG 2.2 AA require, which accessibility criteria an AI design agent can compute, where automated tools stop, and why upstream constraints beat overlays.
Contrast as a Computable Rule: What Breaks When an Agent Swaps WCAG Ratios for APCA
A validated APCA implementation ran against WCAG 2.x on 15 real design-system color pairs. The divergence is polarity-asymmetric — and dark mode takes the damage.
Named but Not Known: Measuring the Gap Between Field Names and Machine Purpose
A static audit separating a field's programmatic name from its declared machine purpose, plus two gates agents can compute from server-sent markup.
The Hover Blind Spot: What Web Agents Cannot Perceive by Default
Nine production pages audited at rest: 1,515 :hover rules against 759 :focus rules. The web's densest interactivity signal channel is invisible to agents by default.
Script-Aware Typography: 4 Rules Design Agents Need to Stop Breaking Non-Latin Text
Western-trained design agents clip Devanagari matras, collide Arabic diacritics, and never wrap CJK. Four computable per-script rules turn typography into lint.
Color Oracle Review: Color Blindness as a Computable Rule
Color Oracle is a free color blindness simulator for Mac, Windows, and Linux. This review explains how AI agents can measure contrast on simulated colors.
What an AI Design Agent Can Perceive About RTL (Arabic) Web Design — Direction as a Computable Design Signal
Most AI design agents assume left-to-right, Latin-script layouts. This review maps the machine-readable signals of Arabic RTL design — dir attributes, logical vs physical CSS properties, connected-script typography — into one computable rule an agent can enforce with a single stylesheet pass.
Axe Accessibility Testing Tool: A 2026 Design-Agent Review
Is Axe the best automated accessibility testing tool? This review covers what Axe does well, where it falls short, and how AI design agents can learn from it.
The 94.8% That Proves Design Quality Can Be Automated
94.8% of websites fail basic accessibility standards and ADA lawsuits hit 5,114 in 2025. For AI design agents, accessibility is the one domain where quality is specified, computable, and legally enforceable — the difference between a pre-deploy lint and a runtime audit.
Leonardo Review 2026 — What an Agent Learns From Contrast-Ratio Color Generation
Leonardo is Adobe's open-source color tool that generates palettes from target contrast ratios instead of manual picking — the closest thing to how an AI agent would think about color.
Why Dark Mode for Developer Blogs?
Developer-focused blogs overwhelmingly use dark themes. Is this just preference or is there a real usability argument? Here's what the research and our own 9-blog empire suggests.