Relative luminance

Relative luminance is how much light a colour puts out, normalised so black is 0 and white is 1. WCAG computes it as Y = 0.2126 R + 0.7152 G + 0.0722 B, and the three weights are the whole story: green carries seventy percent of the answer, red a fifth, blue almost nothing. That is not an aesthetic judgement, it is the sensitivity of human cone response measured in 1931 and standardised since. It is also why pure blue text on black measures almost no contrast while pure green on black measures a great deal, even though both are “a saturated hue on black”.

The linearisation step in front of that sum is the part people skip. An sRGB channel value is not proportional to light; it carries a transfer curve of roughly the 2.2 power, so each channel has to be undone before it can be weighted. This is why #808080 measures 0.216 rather than 0.5. Half the channel value is nowhere near half the light, and any arithmetic done on raw hex values, averaging two colours, taking a midpoint, guessing a contrast, is arithmetic on the wrong numbers.

Everything downstream consumes this one figure. A contrast ratio is (L1 + 0.05) / (L2 + 0.05) with the lighter colour on top, which is where the names L1 and L2 come from; the 0.05 is a constant standing in for ambient light reflecting off the screen, and it is what keeps black against black at 1:1 instead of dividing by zero. Because the formula sees only luminance, two colours can be wildly different in hue and still measure close to 1:1, which is exactly the failure a designer’s eye forgives.

It is worth keeping this apart from lightness, because the alias “perceived brightness” points at the wrong one. Relative luminance is physics weighted by the eye’s sensitivity to wavelength: a linear measure of light. L* in CIELAB, and L in OKLCH, are that same quantity rescaled so that equal numeric steps look like equal steps, which is a perceptual measure. Mid grey sits at 0.216 in luminance and at 53.6 in L*, and both are correct answers to different questions. Contrast maths wants the first; colour ramps want the second. APCA starts from a luminance of its own with a different curve and a clamp near black, for exactly this reason.