ML-backed automatic classification cannot solve 100% of problem, but it can do 90%-95% and left only tenative cases to humans. This combined with an easy human-

ML-backed automatic classification cannot solve 100% of problem, but it can do 90%-95% and left only tenative cases to humans. This combined with an easy human-reviewed appeal system and internal carma mechanics (even in the easiest form of reports/strikes/appeals/lost appeals stats) will reduce the problem 10-20-fold.