Contrast fails under color-blindness simulation

The same text–background pairs as in COLOR-01 are recomputed through a colour-blindness simulation (the colours are converted to how a person with a colour vision deficiency sees them) and the contrast is measured again. The finding covers pairs that fail in the simulation.

So these are pairs that pass or sit just above the line in ordinary vision but drop below it after simulation. The report gives the count.

What the check does not do: it does not simulate every type of colour vision deficiency with equal accuracy, it does not judge whether the text matters to the page, and, like COLOR-01, it takes the background colour as the browser reports it, so for text on a photograph the result is indicative only. The finding is page-level; severity is warning.

For visibility in AI answers there is no effect. And unlike COLOR-01, here we must be careful even with the claim about the standard.

WCAG 1.4.3 does not require a colour-blindness simulation. The standard sets a contrast ratio computed from luminance, and luminance changes little under a colour vision deficiency. This finding therefore goes beyond the standard. It is our own stricter check, not a requirement.

Why we do it anyway: because a colour vision deficiency affects a non-trivial share of the population, predominantly male, and pairs of colours that merely blur a little in ordinary vision can merge entirely for them. Typically green and red, or blue and purple.

How to read this finding: not as a rule violation, but as a warning that your design leans on colour more than it should. When it coincides with COLOR-03 (information conveyed by colour alone) it is a strong signal; on its own it is a prompt to think.

A general rule that resolves most cases: do not build distinction on hue, build it on luminance. Two colours with markedly different lightness stay distinguishable under simulation; two equally light colours of different hue do not.

• Check green and red side by side. It is the commonest pair in designs (success/error) and the most problematic.
• Add a second signal to status colours: an icon, a shape, a word. That also fixes COLOR-03.
• Darken or lighten one of the pair rather than changing the hue.
• Check it yourself. Browsers have colour vision deficiency simulation in developer tools. Switch it on and look at the page through eyes that cannot tell the colours apart.

And a note on sequencing: fix COLOR-01 first. When the contrast fails even for ordinary vision, the simulation only confirms it and fixing twice is pointless.