A blog post describing an idea I had a while ago on how to (sort of) break an impossibility result and achieve asynchronous consensus tolerating <n/2 Byzantine faults, rather than the standard <n/3. emacbrough.wordpress.com/2018/10/24/syb…
P.S., before someone brings it up without having read the blog post, I'll clarify with "modulo standard ways of getting around FLP". You still need either randomization or partial synchrony. The mechanism in this blog post only gets around the n>=3f+1 result.
By the way, in case people found the post confusing, here's a conversation I just had with @ercwl that might help to clarify the key point. 
