ETHPrague 2026

Ethereum's finality today takes minutes. Proposed replacements compress this to seconds, but most designs inherit the classical 3f+1 model and its fundamental trade-offs: either a single leader bottleneck, or multiple rounds of voting, or reliance on synchrony assumptions that are hard to justify in practice. We explore a different region of the design space by working in the 5f+1 fault model, which, at the cost of a moderately larger validator set relative to the tolerated adversary, enables protocol structures that are qualitatively unavailable at 3f+1.
Concretely, we present a family of protocols spanning three complementary points: (i) a parallel leader-and-leaderless consensus protocol that achieves optimistic responsiveness alongside guaranteed prefix progress in every view, even under asynchrony; (ii) the first asynchronous validated agreement protocol in the 5f+1 model, with expected worst-case latency of 3.75 message delays; and (iii) a leaderless, one-round finality gadget that can finalize prefixes of a dynamically available chain without requiring a designated proposer or timeout in the optimistic case.
All three designs share a common philosophy: exploit the richer quorum structure of 5f+1 to decouple finality from leader election, reduce tail latency, and compose cleanly with existing data availability layers. We discuss their trade-offs, how they relate to each other, and what they mean for the practical path toward sub-second Ethereum finality and accountable safety.