As of SERVER-91707, we overwrite all documents' _id fields when inserting them into event collections. But since we do this after sharding is handled, we create orphans. We thus need to overwrite at the router level for sharding to work, and also ensure we aren't then overwriting again later on when we insert into the record store.