Descriptor: Blockchain Data Trust Benchmark Dataset (BDT-Benchmark)

Trust is a foundational principle in blockchain systems, yet the technology does not guarantee the trustworthiness of the data it stores. While blockchain ensures integrity, immutability, and secure consensus, it stores information exactly as received, regardless of its reliability or origin. This limitation introduces risks in decentralized applications, where untrusted data can compromise system behavior and decision-making. Despite the critical need for trust assessment, existing datasets remain limited in scope, often focusing on isolated issues like fraud or smart contract vulnerabilities. To address this gap, we introduce BDT-Benchmark, a comprehensive synthetic dataset designed for machine learning–based trust evaluation across blockchain-integrated environments. BDT-Benchmark spans three core trust dimensions: data source–related, blockchain-internal, and external entity and cross-chain–related features. It includes both benign and malicious records, with representative attacks such as Sybil identities, smart contract exploits, and oracle manipulation modeled using domain-informed statistical distributions. Ambiguous benign data is weakly labeled using Snorkel to scale annotation beyond rule-based heuristics. The dataset is also designed to be extensible, allowing integration with external data-related features from real-world or synthetic sources, enabling broader applicability across domains. We validate the dataset through inter-feature correlation preservation visualizations and machine learning evaluations. Standard classifiers—including Logistic Regression, Random Forest, Naive Bayes and Decision Tree—achieve strong performance across both balanced and imbalanced trust scenarios. BDT-Benchmark serves as a reproducible, extensible resource to benchmark trust assessment models and can support research in secure blockchain integration with IoT, DeFi, and decentralized oracles.