All documentation about this PAWS bot run is attached to the notebook. Note that the page will take some time to load because there’s a lot of bot output.

If you don’t want to wait for that page to load, here’s the short version: Many German cities had many population statements with timestamps, which all had the “normal” rank, so a simple WDQS query with wdt: would return all of them. The bot set the latest one to “preferred” if the latest and second-latest one had exactly the same source.