As long as the "100 million dollars" don't make up more than 50% of the tokens in the network, the incentives are supposed to play out in a way that all the oth

As long as the "100 million dollars" don't make up more than 50% of the tokens in the network, the incentives are supposed to play out in a way that all the other participants of the network contest the wrong statement made by B, and thus B isn't able to introduce information that is deemed wrong. So in general the situation should be preventable. There are some specific versions of that edge case that come to my mind that where that situation could indeed be unresolvable but in those cases that should be detectable and displayable to the users of the system. As I said in my previous comment, there are some rationality requirements regarding some of the token amount.