we had an implementation bug where we weren't calling identify properly on a subset of users, meaning that the events of those sessions were recorded as anonymous, and not properly connected to their user profile; however, some of those events held enough information in their properties for us to determine who the user was, and retroactively identify them... we really needed only one such event per anonymous id; instead we ended up downloading all those events, which were quite numerous, and then filtering them down to one on our end