Now in production on Aptos, significantly cutting latency. The core idea — Prefix Consensus: replicas vote on the longest prefix of data they have, not the full block, avoiding fetching on the critical path.
Robustness + speed, no trade-off. The full paper by Andrei Tonkikh, Balaji Arun, Daniel Xiang, Zekun Li and Alexander Spiegelman breaks down how Prefix Consensus works.
Read it here: https://arxiv.org/abs/2504.18649
