She and one other researcher has manually classified all 7000 papers as per standard protocol. Perhaps for the next article they will measure how this tool agreed with them against them and include it in the protocol if good enough.