I have read a description of a Quantum eraser experiment: http://www.bottomlayer.com/bottom/kim-scully/kim-scully-web.htm I don't understand the technicalities, like the process how the entangled photon pair is generated. So things like that may solve my problem, or may not. What I'm not getting is this: why do we need to include both a part with "which path" information and one without, simultaneously in the same experiment. This makes the final correlated information seem like a sampling of the distribution at D0. Whereas, making the idler side always erase path information (or always keep it) would show a 100% picture. It seems reasonable to assume, that this simpler experiment has been tested before inventing the one in the link, with that labyrinth of mirrors and splitters and coincider circuit. Based on what I understand, I would expect the following to happen (which Im sure is wrong somehow): If we just remove the idler side of the experiment completely, we should not see any interference pattern, due to the path information being observable (?). Even if Im wrong about this, we can then put there detectors that tell us the path information, no problem here. However, if we put the eraser in the idler side, it removes the path, and the interference pattern should appear at D0. It worked when applied to only D3 and D4, so why not now? I understand that in the original experiment there was no interference at D0, somehow I feel that should be important for this? If this actually worked, it would mean that my choice of setting up the device determines if there will be path information. However, that in turn determines the distribution pattern at D0. Just for visibility, lets say that the idler part of the experiment and D0 detector are in different rooms (they could be galaxies apart), then me changing the setup in one room would change the interference pattern in the other room. Surely this cant be true. Not because it would be counterintuitive, Im ready to accept that. But this would be a relatively simple experiment (compared to the one in the link), and it would completely disprove local hidden variables, much more obviously than Bell's inequality, making that obsolete. So the pure existence of Bell's theorem indicates that it wont work like I described. But then how?