PRECISE: Precision-Driven Discovery of Token-centric MEV

Abstract

Token-centric maximal extractable value (tMEV) arises when token supply-control behavior interacts with exchanges whose pricing ignores particular balance updates. Existing analyses face state explosion, missing deep inter-transaction paths or producing false positives. PRECISE combines token-level static filtering, fork-based dynamic fuzzing, and state-difference verification. Inter-procedural system dependence graphs (SDGs) identify candidate paths from source or bytecode; utility-guided fuzzing validates them on a pinned Anvil mainnet fork against authorization-, event-, and profitability-aware Type-1/Type-2 predicates. This division limits dynamic search to supply-sensitive control flow while retaining execution-grounded validation. Profitability follows a constant-product automated market maker (AMM) slippage model. PRECISE scans 22,279 token contracts and 7,030 pools; on the 499-contract standard scored set it reaches 72.45% precision, 84.58% recall, and 78.04% F1 against Nyx, FORAY, DeFiTainter, and SmartCAT. Recall rises to 91.2% on a 150-contract Complex-Tokens stress set. The artifact pins replay contexts and reproduces the scan for an estimated compute cost of $25.84 at the reported instance-hour rate.