On-Chain Reputation and Anomaly Scoring: A Cross-EVM Risk Engine for Sybil Detection, Wash-Trading Identification, and Marketplace Trust Operationalisation

Abstract

Decentralised marketing platforms that operate across multiple EVM-compatible blockchains face a distinctive risk surface: the same adversarial patterns observed in decentralised finance (sybil networks, wash-trading rings, bridge-based provenance laundering) migrate directly into influencer campaign ecosystems, where inflated wallet activity translates into inflated audience claims and fraudulent escrow eligibility. Existing blockchain analytics tools provide raw attribution data but stop short of operationalising their signals into real-time marketplace controls. This paper presents the Cross-EVM Reputation and Anomaly Scoring Engine (CRASE), a distributed analytics architecture that normalises on-chain telemetry across Ethereum, Polygon, Arbitrum, Base, and Solana, computes composite wallet reputation scores, and feeds actionable risk verdicts into both discovery ranking and escrow gating logic. CRASE employs a fourcomponent scoring model: transaction graph topology analysis for sybil cluster detection, temporal pattern analysis for wash-trading identification, cross-chain provenance tracing for bridge-based obfuscation detection, and wallet-age and activity normalisation for cold-start equity. Evaluation against a labelled dataset of 120,000 wallets drawn from publicly documented DeFi exploit post-mortems and influencer fraud reports yields a sybil detection F1 of 0.88 and a wash-trading detection F1 of 0.83, with score computation latency below 850 ms at the 99th percentile. Critically, CRASE outputs are consumed directly by marketplace discovery and preescrow settlement controls, closing the gap between blockchain signal collection and productlevel risk enforcement that characterises less mature platform architectures.