IMO, detection/observation is the wrong framing of the problem. A better approach is about consistency. All interactions in a system must be consistent: if you

IMO, detection/observation is the wrong framing of the problem. A better approach is about consistency. All interactions in a system must be consistent: if you have two Schrodinger's cat-boxes being controlled by the same measurement, then either both cats will be dead, or both will be alive, you will never find one dead and the other alive. In an experimental setup, you might have many different levels of interaction: the photon either hits one sensor or another, that results in one or another message being sent to your recording device, which records one or the other result on your harddrive, etc. As the scale goes up, more and more stuff gets dependent on the result of your measurement, until eventually everything is, yourself included. And if everything is dependent on the result of your experiment, they must all be consistent with a given result of your experiment. Thus, you end up with exactly one answer. Since the slit edge isn't changed by the diffraction of light, there is no inconsistency between whether the photon passed through its slit or the other.