論文題目： Byzantine Fault-Tolerant Consensus （與陽明交大電機系共同指導）
論文摘要：
Reputation-based primary selection has been widely adopted in PBFT variants to accelerate recovery from Byzantine primaries. However, the correctness implications of relying on locally computed reputation scores have not been formally studied. In this paper, we identify and formalize a fundamental vulnerability, called {\it Primary Selection Failure}, which arises when honest replicas maintain inconsistent local reputation scores and therefore select different primaries during a view change. We show that primary selection failure can manifest in two forms. In collision, multiple replicas each collect at least $2f+1$ \texttt{VIEW-CHANGE} messages and simultaneously believe they are the new primary. In livelock, no replica reaches the required threshold and thus no primary can be established. We further show that, in reputation-based PBFT with inconsistent local scores, collision can threaten safety by enabling conflicting prepared certificates, while livelock can violate liveness by preventing progress. To quantify the severity of this problem, we study a random primary selection baseline in which honest replicas independently choose primary candidates uniformly at random, a maximum-divergence reference case for inconsistent reputation views. Under this model, we derive exact probabilities of collision, success, and livelock for arbitrary $n$ and $f$, and develop an efficient recursive dynamic programming method for their computation. We also show that collision is impossible when $n < 5f + 2$. These results reveal a fundamental correctness risk in reputation-based PBFT and characterize the conditions under which different primary-selection failure modes emerge.

已發表/已接受

[會議論文]

W.A. Prabowo, S.W. Wang, S.S. Tzeng, and L.C. Wang, "Collision and Livelock: Primary Selection Failures in Reputation-Based PBFT with Inconsistent Scores," in 2026 IEEE Global Communication Conference (Globecom 2026), Macau S.A.R., China, December 7-11, 2026.

---

準備中/修改中/審查中

[審查中]

W.A. Prabowo, S.W. Wang, S.H. Cheng, and L.C. Wang, "SHIELD-MCPS: Intelligent, Fair, and Adaptive Leader Selection for Byzantine Fault-Tolerant Consensus in Permissioned Blockchains," for IEEE Transactions on Network Science and Engineering, (Last Update: 2026-08-22 Revising after Revise and Resubmit)