A serious solution is simply for several major stakeholders to publish signed endorsements and/or kills to blocks. Users can subscribe to them, and if a supermajority agrees to kill a block, the merchant at the very least can be configured to stop and go into safe mode.
I have thought of this long ago, now others might take the idea seriously. It was aggressively rejected in the past under the pretense it was too "centralized". I believe we need it to raise the bar on the risk of a 51% attack.
Your centralized protocol with few designated major stakeholders is indeed a terrible idea imho, for the reasons that we discussed in your thread. Fully decentralized proof-of-stake protocol can be sound, and can have several additional features that are very beneficial (in exchange for the additional complexity in the work that the nodes perform), but the security analysis should take into account bribe attacks etc., as discussed in the PoA thread. Peer review for the best protocol that we came up with so far would be very welcome, see
here.