Web of trust seems somewhat risky in hiding valid replies. Especially with the significant prior work Terry did on translation support, I wonder whether he sees a machine learning model angle to filter nonsense or generic content. Essentially this seems to boil down towards an AI content detection issue.