I wrote briefly about my relative qualifications at the following link:
https://bitcointalk.org/index.php?topic=226033.msg2441624#msg2441624I will now address all discussion since my prior post in this thread.
1. I thought we already decided that excluding transactions was not a possible attack (unless 100% of peers are evil), because the honest fork would be readily identified as including all transactions (including those from the evil fork). The worst that could happen is a transaction could be delayed by one CB period, if the evil fork withheld propagation of TBs (to non-evil peers) until the CB (where it can't be withheld any more and still be a public fork) or if the evil fork included transactions from the honest fork one CB later (so as to be not readily identified as the evil fork).
This delay could allow for a double-spend, where conflicting forks have different order of which of the double-spend transactions was first.
2. The only attack vector is that evil peers will not let non-evil peers sign their CBs and evil peers could create one or more forks in addition to the honest peers. There might even be multiple factions of evil peers who don't want to share tx revenue and thus exclude other peers. So the problem is how to identify which fork is the consensus in order to properly award tx fees?
In light of the above, the proposal since my prior post appears to distill to that peers who attempt to sign a CB and are ignored by a fork, will see that fork as dishonest if propagation error/delay can be ruled out.
The problems are:
a. Evil forks might occasionally allow other peers to sign, thus obfuscating whether propagation or evilness is the cause of missed CB signatures.
b. The honest peer's opinion is only valid for himself, and can't be proven to the world without 51% consensus.
No matter how you slice it, you can not avoid the
51% Rule of Decentralized Agreement.
This is not a crippling conclusion. We should move forward on improving upon Bitcoin. We will not be able to make a decentralized currency that avoids the 51% attack.