Outils couleur

Hex for a value you type once and never touch again — it is compact, universally understood and lossless. HSL when a human is going to edit it by hand, because 'a bit darker' and 'less saturated' are single-number changes there. OKLCH when a program computes the colour: a generated scale, a gradient, an interpolated hover state. OKLCH is the only one of the three whose lightness number matches what the eye reports, so evenly spaced numbers come out looking evenly spaced.

Almost always rounding rather than disagreement. HSL as normally written — a whole-number hue and a tenth of a percent — holds fewer distinct values than the 16.7 million a hex can express, so several hex values collapse onto the same HSL and the conversion back lands on whichever the formula produces. Where tools genuinely differ is in how they report a grey: the hue of a colour with zero chroma is undefined, and some print 0 while others print whatever the arithmetic produced.

Yes. It is the WCAG 2.x formula: linearise each channel with the sRGB piecewise transfer function, weight them 0.2126, 0.7152 and 0.0722, add 0.05 to both luminances and divide the larger by the smaller. The most common mistake in a home-made implementation is raising the channels to the power of 2.2 instead — close, but it shifts the ratio enough to flip a borderline pair from pass to fail. The APCA figure shown alongside is a different model entirely and is informative, not a standard you can claim conformance against.

As a starting point, not as a specification. CMYK is device-dependent: the same sRGB colour becomes different ink percentages on coated stock than on uncoated, and different again on a different press with different inks. That is what an ICC profile encodes, and no web page has one for your printer. Send the printer the sRGB or Pantone reference and let them convert with the profile for the job.

It is a screening indication, not a diagnosis. A clinical result comes from an anomaloscope under controlled lighting on a calibrated display, and none of those conditions hold on your desk. What an online test can do honestly is tell you whether you read the same figures as someone with typical colour vision, which is enough to be worth following up on. The plates here are generated from confusion pairs found by simulation rather than scanned from a copyrighted plate set, which removes the single largest error source in the usual online version: a JPEG of a 1917 print, recompressed, at whatever colour accuracy the scanner had.

Because they are generated in OKLCh rather than by stepping HSL lightness. HSL weights all three channels equally when human vision weights green roughly ten times more than blue, so a ramp built on it looks light in the yellows and dark in the blues at identical numbers. Stepping OKLCh lightness on a curve, holding hue, and tapering chroma at both ends produces steps that look as even as they measure — and every stop is clamped back into the sRGB gamut, so none of them render as a flat clipped colour.