All But the Edge
Two surfaces that are the same number, split by a border that repaints both of them. Narrated by an AI that can read the byte at each end and watch you disagree with it.
In The Edge That Isn't you saw a triangle nobody drew. This is that experiment run backwards. The edge here is real, sharp, drawn honestly down the middle. What is not real is the thing everyone reports afterwards: a difference between the two big flat regions on either side of it.
Those regions are one eight-bit value. Not close. The same integer, 128, and the next screen reads it back out of the framebuffer at both ends while you are looking at them.
This is the Craik-O'Brien-Cornsweet effect. Craik worked it out in the 1940s, Vivian O'Brien published the contour version in 1958, Tom Cornsweet put the figure in front of everyone in 1970. Approaching the border from the left the luminance ramps down; from the right it ramps up; past the ramps, nothing differs. Your visual system takes the sign of that border, which is the only real information in the picture, and paints it across both surfaces.
- 0️⃣ The proof: the figure, a live pixel readout at both ends, a strip you can drop over the border, and a button that hides everything except what is physically different.
- 1️⃣ The nulling: seventeen quick judgments. Each one gives the two surfaces a real luminance step against the illusion, and an adaptive staircase hunts the step that exactly cancels it.
- 2️⃣ The fork: eleven more with the same border amplitude smeared over a ramp six times wider. That is more total ink. Two accounts of this effect predict opposite results, and your numbers pick one.
- 🖥 Turn off night-shift, warm-light and auto-brightness if you can. They rescale everything this page measures.
- The border flips polarity at random on every single trial. Press the same button all the way through and you null at zero by construction. A real percept predicts your answer flips with the border, and expectation does not predict a flip.
- Two staircases run interleaved, one climbing from zero, one falling from over-cancelled, so the answer cannot be an artefact of which side we came from.
- Catch trials have no border cusp at all and a real step, so one side genuinely is lighter. Fail them and this page says so instead of scoring you anyway.