Secure Transaction Semantics: Analysis, Vulnerability Detection, and Attack Modeling

Abstract

Blockchain transactions are often interpreted by off-chain systems through call traces, event logs, and storage modifications. However, these artifacts can diverge from the actual on-chain execution due to semantic mismatches caused by reverts or misleading logs. Existing tools largely assume consistency between observable effects and final state, overlooking semantic mismatches. We present a semantic framework for smart contract security analysis that models and leverages transaction-level semantics to detect vulnerabilities, synthesize attacks, and explain off-chain inconsistencies. Our approach identifies mismatches between real execution effects and intent-oblivious interpretations by off-chain systems. We plan to implement three tools: PEvent-Catcher for detecting log forgery vulnerabilities, RollGain for synthesizing rollback-based state-reverting attacks, and DeepTx for real-time intent detection.