Consolidation of Ground Truth Sets for Weakness Detection in Smart Contracts

Smart contracts are small programs on the blockchain that often handle valuable assets. Vulnerabilities in smart contracts can be costly, as time has shown over and over again. Countermeasures are high in demand and include best practice recommendations as well as tools supporting development, program verification, and post-deployment analysis. Many tools focus on detecting the absence or presence of a subset of the known vulnerabilities, delivering results of varying quality. Most comparative tool evaluations resort to selecting a handful of tools and testing them against each other. In the best case, the evaluation is based on a smallish ground truth. For Ethereum, there are commendable efforts by several author groups to manually classify contracts. However, a comprehensive ground truth is still lacking. In this work, we construct a ground truth based on publicly available benchmark sets for Ethereum smart contracts with manually checked ground truth data. We develop a method to unify these sets. Additionally, we devise strategies for matching entries that pertain to the same contract, such that we can determine overlaps and disagreements between the sets and consolidate the disagreements. Finally, we assess the quality of the included ground truth sets. Our work reduces inconsistencies, redundancies, and incompleteness while increasing the number of data points and heterogeneity.