Lab · 9
Can AI make WCAG contrast rules understandable to non-experts?
Experiment 8 built an internal checker that enforces WCAG contrast automatically, before anything ships. This one asks a different question — take that same math out of the pre-deploy gate and put it directly in front of a visitor. Can AI turn a formula almost nobody outside accessibility work knows into something anyone can just try? Try it yourself: the contrast checker.
Disclaimer: the contrast checker is for demonstration purposes only. It isn’t a certified accessibility auditor and shouldn’t be used to ensure accessibility compliance — verify anything shipping to real users with a dedicated accessibility tool.
Hypothesis
Can AI make WCAG contrast rules understandable to non-experts?
Finding
The contrast checker works! It’s interesting to use it and discover what works and what doesn’t work.
Tested by building a live tool that runs the exact same WCAG contrast math as Experiment 8’s internal checker, then actually using it against real color pairs — not a formal usability study, and not tested with people who’d never heard of WCAG before.
Method
Reuse
Ported Experiment 8’s exact contrast math — relative luminance, gamma correction, the WCAG ratio formula — from Python into JavaScript so the two never quietly disagree.
Build
Built swatch pickers and hex inputs for foreground and background, a one-click swap control, and a live text preview at both sizes.
Translate
Turned the four WCAG thresholds — AA and AAA, at Normal and Large text — into plain pass/fail language instead of a bare ratio number.
Ship
Deployed the tool live at /contrast-checker/ for anyone to try, and documented it in the design system.
Tech
The experiment
One live tool, four WCAG thresholds checked at once, and the same math Experiment 8 already proved against the real site.
AI handled
- Porting the exact relative-luminance and contrast-ratio math from Experiment 8’s Python checker into JavaScript so the two implementations can never drift apart
- Building the swatch-plus-hex-input pair for foreground and background, a swap control, and a live text preview at Large and Normal size
- Translating all four WCAG thresholds — AA/AAA at Normal/Large text — into a results panel that reads as a plain pass or fail, not just a raw ratio
- Documenting the tool as a new organism in the design system and adding it to the site’s own accessibility audit script
Dr. Paul handled
- Setting the hypothesis
- Choosing to ship it as a standalone tool rather than only folding the write-up into the Lab
- Giving explicit go-ahead to push the full tool live
- Confirming the finding and status in his own words
Worked: The exact math Experiment 8 already proved against the live site now runs instantly in the browser for any two colors — turning a formula almost nobody outside accessibility work knows into something anyone can just try, and see fail or pass in plain language.
Constraint: The tool checks contrast in isolation. It doesn’t, and can’t, tell anyone whether a color pair actually reads well in context — font weight, real-world size, and the surrounding design still need a human’s eye.
Finding
The contrast checker works! It’s interesting to use it and discover what works and what doesn’t work.
Built on the exact same relative-luminance and contrast-ratio math Experiment 8 already proved against the live site — the tool and the internal checker can never quietly disagree.
Checks all four WCAG thresholds — AA and AAA, at both Normal and Large text — the moment two colors are entered.
Implication
Experiment 8 proved AI could enforce WCAG contrast automatically, quietly, before anything ships. This experiment asked what happens when that same math stops being a gate and becomes something a person can actually touch. The difference turned out to matter: reading a pass/fail report is not the same experience as picking two colors yourself and watching the number move. Dr. Paul’s own reaction — that it’s interesting to use and discover what works and what doesn’t — is itself the finding. An enforcement tool tells you that you were wrong. An exploration tool lets you find out why, on your own terms, at your own pace. Both are legitimate uses of the same formula; they just serve different moments. The rule didn’t change between Experiment 8 and this one. What changed was who it was built for.
The same formula that silently gates a deploy can also become something people want to sit and play with.