You never see reddish-green or yellowish-blue — because the eye doesn’t send ‘red’ and ‘green’ separately. It sends their DIFFERENCE on one wire: red-vs-green, blue-vs-yellow. Stare at red until that wire tires, then look at white, and the tired wire swings the other way — you see green that isn’t there. Slide the staring and watch the afterimage bloom.
Opponent-process theory: three cone types feed two chromatic OPPONENT channels — red−green and blue−yellow — plus a luminance channel. A neutral gray sits at zero on both chromatic channels. Staring at red drives the red−green channel hard, and it ADAPTS (its gain drops on the red side); afterwards a gray input is read as NEGATIVE red−green = green. The instrument runs the opponent transform and the adaptation live. A fail-loud self-check throws unless gray reads zero on red−green, and after red adaptation the same gray reads negative (green) — the afterimage as arithmetic.
A simplified three-channel model with linear adaptation; real opponency is built from noisy cone signals with nonlinear gain control, and afterimage colour depends on the exact adapter. The opponent transform and the sign-flip after adaptation are exact (Hering 1892 / Hurvich-Jameson 1957).