SLID Scam Detection under Temporal Drift: A Practical Framework for DeFi Security †

The Slow Liquidity Drain (SLID) scam has recently become a significant problem for decentralized exchanges (DEXs) and the decentralized finance (DeFi) environment in general. As SLID evolved from rug-pull scams toward cryptocurrency transaction fraud, several studies have implemented and published methods of detecting SLID scams, including heuristic-based and machine-learning (ML) based techniques. However, the applications of those detection methods to the real running DeFi application on blockchain are limited due to several constraints. Specifically, the newly arrived datasets that were collected from recent DEX activities have revealed that SLID behavior could evolve, which leads to the obsolescence and ineffectiveness of the previous methods when applying static thresholds to the detection in a real-time system. In this paper, we present a data-driven revalidation of SLID detection under various DeFi conditions and discover a robust method for threshold and feature-importance updating as the SLID behavior evolves over time with new data emerging. From the findings, we propose two practical and applicable models: A slow-adaptive solution that leans toward stable detection over a long period of time, and a fast-adaptive solution for more robust and real-time-sensitive detection. Those models together transform the theoretical detection methods to adapt to the real-world system that requires robustness and adaptability to various SLID evolutions.