As you probably know, occasionally when QuickStatements is running merge commands, it will sometimes empty an item, but then fail to redirect it. Usually such cases are picked up by PLbot within a few hours, which completes the redirect.
Recently I've been using QS to do quite a few merges involving items from the import from ThePeerage, which can include a number of parent/child or spouse/spouse relationships.
Usually everything goes fine. However, User:From Hill To Shore has found that in at least one case after QuickStatements had done an incomplete merge of an item, which had a parent/child statement, that before PLbot could complete the merge and put in the redirect, because the father (P22) statement on the child item was still pointing to the original item, and the original item wasn't identified as a redirect, DeltaBot moved in and restored the inverse child (P40) statement on the original item; which was then no longer empty, so PLbot could no longer identify the redirect needing to be completed.
This was one case of this happening that was identified, but I have no idea whether this race-condition may have been widespread, and whether there may have been more. Also, with the items now so emptied, apart from the restored statement, other cases where this may have happened may be quite hard to identify. I'd be grateful if you have any ideas? But going forward, it might be worth DeltaBot checking to see whether there is a recent "Merged item into" edit summary on the item, before restoring any inverse statements.