Are 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.

EDIT: 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)