Design and Evaluation of State-Based Testing Frameworks for Blockchain Consensus Protocols

Blockchain technology enables decentralized consensus-based systems that elim- inate centralized trust, with consensus protocols serving as the core mechanism for agreement on a shared system state. The security and reliability of blockchain systems critically depend on the correctness of these protocol implementations. However, testing consensus protocols poses unique challenges that traditional software testing techniques fail to address adequately. Existing consensus testing frameworks suffer from several fundamental limitations. Static state models derived from normal protocol execution overlook rare or unexpected behaviors. Network-level instrumentation exposes only message exchanges and does not provide visibility into internal node states or subtle logical flaws. Random or queue-based mutation strategies explore the large and complex consensus state space inefficiently. As a result, critical vulnerabilities may remain undetected, leading to consensus failures, chain splits, and financial losses. This thesis presents a comprehensive blockchain consensus testing framework addressing these limitations through three complementary contributions. First, AWOSE introduces probabilistic state modeling based on Markov Chains and Hidden Markov Models to capture dynamic consensus behavior. Second, UNBUG proposes a multi-dimensional bug detection framework combining Kalman Filters and Hidden Markov Models to identify memory, semantic, concurrency, and environmental bugs. By integrating node-level instrumentation with network-level monitoring, UNBUG enables detection of subtle logical and timing-dependent issues beyond crashes and memory corruption. Third, COMPOTE introduces a context-aware message prioritization and mutation framework that leverages selective parsing, feature extraction, and DBSCAN-based clustering to guide efficient consensus fuzzing. Together, these contributions advance the state of the art in blockchain consensus testing by combining probabilistic modeling, multi-dimensional bug detection, and intelligent test case prioritization.