Personally I try to align new POIs that I cannot see on aerial imagery as well as I can to existing geometry like building outlines or streets/paths. Either one

Personally I try to align new POIs that I cannot see on aerial imagery as well as I can to existing geometry like building outlines or streets/paths. Either one of those may we wrong since aerial images often have an offset that varies depending on where you are and you can't really tell from the map which points are accurate and which ones aren't. So your bench that's accurate to 10 cm may just as well be moved by someone else again because it's "clearly near that path intersection and not 10 m away from it". Maps basically lie all the time and every map is an imperfect approximation of reality. Of course if your envisioned data consumers are blind people that are guided by surveying GPS equipment to that bench so they can just sit and be confident in actually sitting on the bench ... well, yeah, that might work. But I don't think that's a particular compelling use case and most people when they're looking for the bench that's on their map will find it just fine if it's not that accurate. Interesting tidbit: Hiking guideposts around here all have UTM coordinates written on them and a few of them are a bit wrong, at least compared with other geometry sources that should in theory also be accurate (like cadastre data).