I have been mulling over various aspects of delayed quantum erasure and came upon the following puzzle. It is about the famous 'delayed choice quantum eraser' of Kim, Kulik, Shih and Scully. The paper says (1st para of 2nd column on p2) that the path length from the BBO crystal that generates the entangled pair to the detectors D1, D2, D3 of the 'idler' photon is 2.5m longer than the path from BBO to D0, the detector of the 'signal' photon (from which interference fringes may/may not be observed). This equates to a delay of 8ns, which is substantially longer than the 1ns response time of the detectors. It occurred to me though, on looking at the diagram (bottom left of last page), that the 'choice' to erase or not erase is not made at the detectors D1-D3 but at the beam splitter BSA. That is what determines whether a photon is deflected downwards to D3, thereby preserving which-path info, or transmitted on towards mirror MA. After MA. As soon as a photon is transmitted, it becomes inevitable that it will be registered by either detector D2 or D3, neither of which can discern which-path info, so the erasure has happened as soon as the photon clears BSA. So the 'choice' to erase or not erase has been made as soon as the photon has traversed BSA. That choice is only 'delayed' if the path length from BBO to BSA is greater than from BBO to D0. The authors do not report that path length. I know the diagram is not to scale but on the diagram it looks shorter. If that analysis is right then I can see no good reason to regard the erasure as delayed. Agree / disagree? If the analysis is right then the natural response would be to perform an experiment in which the path BBO to BSA is longer than from BBO to D0 by the same sort of margin (about 2.5m). Have such experiments been done? If not, is that because there is a practical difficulty in doing them? Looking at the diagram I can't see any reason why there should be a practical difficulty. Here's a link to a diagram of the experiment on wikipedia. The setup in that diagram has one extra detector D4 of photons carrying which-path info, compared to Kim's diagram, but D0-D3 all denote the same detectors in both diagrams, as do BBO and BSA (BSa).