How do you catch stuff that validates but still fails the receiver’s security model? This paper calls it TBSGs and splits it 4 ways.