You have a lot of context/metadata like date taken (or at the worst, date acquired for a historical collection, giving a terminus ante quem) and can easily figu

You have a lot of context/metadata like date taken (or at the worst, date acquired for a historical collection, giving a terminus ante quem) and can easily figure out the false positive rate. You may not know if two matches are the same person, but if one photo is dated to 1850-1860 and another photo is dated to 1931, you can safely infer that you have a false positive. Over a dataset of matches, that gives you an idea of how reliable each one. Give people access to all the other clues like "insignia and inscription" and probably a lot of reliable identifications can be made, and the remainder flagged as 'possible but uncertain'.