A multi-source empirical audit of blockchain and DeFi: concentration, intelligent agents, and mechanism design

Abstract

Blockchain, decentralized finance (DeFi), and artificial intelligence are converging into an economic ecosystem whose concentration, risk, and mechanism design have rarely been examined within a unified empirical framework. This study integrates five publicly available data sources: Uniswap decentralization indices, the Elliptic Bitcoin dataset, a multi-year sample of ERC-20 trading activity, Bitcoin incentive series, and DeFiLlama Total Value Locked (TVL) data. We processed four sources through a common audit and canonicalization protocol, while adopting the Uniswap decentralization indices as published derived series and harmonizing them within the same analytical framework. The results reveal extreme inequality in the protocol-level distribution of Total Value Locked (Gini = 0.985 for aggregated TVL; Gini excluding CEX = 0.978 ), although the corresponding Herfindahl–Hirschman Index does not indicate high protocol-level market concentration under conventional thresholds. A Random Forest classifier identifies illicit Bitcoin transactions with a precision of 0.988 and a recall of 0.692, consistent with the original Elliptic benchmark. SHAP analysis assigns 74.4 % of the aggregate mean absolute attribution to transaction-level features and 25.6 % to neighborhood features. In addition, Aave V3 exhibits a higher borrowed-to-deposited ratio than Compound V3 (0.410 vs. 0.300), while the observed utilization–volatility associations do not establish a uniform relationship across protocols. Taken together, these findings show that inequality, risk, and mechanism-design characteristics remain jointly observable across the blockchain and DeFi systems examined. Beyond the empirical findings, this study introduces a reproducible audit protocol for evaluating blockchain-based economic systems through explicit cross-source verification, result traceability, and scope-calibrated inference.