Thus if you claim it is not solvable, then you have either found an error in their proof, or you have redefined the problem in your own way. I suspect the latter.
As with all such systems, fault tolerance is achieved up to a specified number of faults, and no farther.
If messages of the generals can't be faked then even 99 of 100 traitors is not a problem. I think he didn't redefine the problem, more likely he just forgot to provide some details.
EDIT:
From that paper:
With unforgeable written messages, the problem is solvable for any number of generals and possible traitors.
No, he mentioned that in his reply, as I did earlier.
.e.g as you say "unless you add externally-assigned identities and unforgeable messages".
He just fails to acknowledge that "Byzantine fault tolerance" only succeeds up to a threshold of faults.
"Correctly functioning components of a Byzantine fault tolerant system will be able to provide the system's service, assuming there are not too many faulty components"
https://en.wikipedia.org/wiki/Byzantine_fault_toleranceYou can specify a high threshold but that greatly constrains the available solutions (in my opinion to largely if not entirely useless ones in the context of cryptocurrencies, but not everyone necessarily agrees). Arguably a low threshold is also largely useless (it seems somewhat fewer, at least within the cryptocurrency community, agree with this statement), which would mean there are no very useful cryptocurrencies possible. That would not surprise me much.