- For attestors (validators who make attestations), only the attestations that are included in a descendant block of B are counted.
The problem with this approach is that it removes a lot of the important benefits that we gain from counting attestations outside of blocks. Particularly, it hands local-consensus-making power back into the hands of proposers, and so gives short-range attackers power to do nasty things. One example is that if you control two slots in a row, and the head before you is B, then you can make a sister of B (call it B’), attest to it, make a child B’’ including the attestations, and then from the point of view of the attesters in the slot after B’‘, B’’ is the winner, and after their attestations supporting B’’ it wins the fork choice counting all attestations too.
The fact that the current LMD design allows for hundreds of independent “confirmations” to happen in parallel is a key part of its power to avoid such situations.