SingularityBFT: Fast Finality and Continuous Asynchronous Recovery

2027 · 2027 Competition

School: School of Computer and Information Sciences
Category: Computer Security and TechnologyPrimary

Project Overview

One Liner: A Byzantine fault-tolerant consensus protocol that combines fast-path finality with continuously available asynchronous recovery while preserving safety across delayed and late-arriving certificates.

Abstract

Modern Byzantine fault-tolerant consensus protocols often combine a fast path for favorable network conditions with a slower recovery mechanism for periods of delay or failure. A difficult edge case occurs when the system begins recovery while enough previously issued votes still exist for an earlier proposal to become final later.

SingularityBFT investigates how fast finality and asynchronous recovery can safely coexist under this condition. The protocol separates transaction-data availability, deterministic ordering, fast-path certification, and recovery into explicit interfaces. Its River fast path certifies exact successor states, while a continuously operating DAG-based ordering layer, Ocean, coordinates recovery when the fast path stalls. Parent-local closure rules preserve any earlier decision that could still become final while preventing recovery from creating a conflicting successor.

The project includes a formal protocol specification, quorum and safety arguments, an executable signed reference implementation, adversarial regression tests, and experimental evaluation of dissemination and recovery behavior. The current reference suite tests late certificates, competing recovery records, persistent signing restrictions, crash-recovery behavior, and deterministic ordering. The senior project will extend this work toward an integrated implementation and end-to-end evaluation under realistic network conditions, measuring finality latency, transaction inclusion, recovery behavior, resource costs, and consistency across failures and restarts.

No video available.

Screenshots

0 image(s)

No screenshots uploaded yet.

Team Members

Naman Bajpai
Lead

Advisors

Brian Stuart