It's worse.
replyThe code contains a function that, given the target ΔE, generates two colors in floating-point Oklab representation, separated by that distance. But there is no check whether the two generated colors end up rounding to exactly the same one on 8-bit displays. So, I was asked to find a boundary (while the claim was that there were two distinct colors 0.0013 ΔE apart) between RGB(80, 83, 152) and RGB(80, 83, 152). Obviously unfair.
I will get around to fixing this. An oversight. Apologies.
replyAre you using Oklab channels to measure delta-E / difference? If so, Oklab is a hacky way to approximate a real colorspace with just one matrix multiplication, the channels have no meaning and are not related to delta-E. Use real Lab*, it'll take 10 minutes with an LLM.
replyEDIT: Just read the blog post. I thought HSL was bad for design, Oklab is much worse. It just goes right ahead and reuses color science terms so it sounds it got it all right. (dEOK existing and its "JND" being 0.02 absolutely made my head spin. None of this is recognizable to a color scientist)