Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MATEX: A Multi-Agent Framework for Explaining Ethereum Transactions

Zifan Peng, Jingyi Zheng, Yule Liu, Huaiyu Jia, Qiming Ye, Jingyu Liu, Xufeng Yang, Mingchen Li, Qingyuan Gong, Xuechao Wang, Xinlei He

TL;DR

The paper tackles the difficulty of understanding complex DeFi Ethereum transactions and the risk of blind signing. It introduces TxSum as a faithfulness-driven NLP task and TxExplain as a real-world dataset of 100 annotated transactions, both grounded strictly in on-chain evidence. It then presents MATEX, a multi-agent framework with a shared Evidence Board and four specialized agents (Profiler, Protocol Investigator, Synthesizer, Safety Auditor) to generate attributed, verifiable narratives and validate them adversarially. Across evaluations, MATEX outperforms baselines in accuracy, faithfulness, and risk identification, underscoring its practical potential for safer, clearer DeFi interactions.

Abstract

Understanding a complicated Ethereum transaction remains challenging: multi-hop token flows, nested contract calls, and opaque execution paths routinely lead users to blind signing. Based on interviews with everyday users, developers, and auditors, we identify the need for faithful, step-wise explanations grounded in both on-chain evidence and real-world protocol semantics. To meet this need, we introduce (matex, a cognitive multi-agent framework that models transaction understanding as a collaborative investigation-combining rapid hypothesis generation, dynamic off-chain knowledge retrieval, evidence-aware synthesis, and adversarial validation to produce faithful explanations.

MATEX: A Multi-Agent Framework for Explaining Ethereum Transactions

TL;DR

The paper tackles the difficulty of understanding complex DeFi Ethereum transactions and the risk of blind signing. It introduces TxSum as a faithfulness-driven NLP task and TxExplain as a real-world dataset of 100 annotated transactions, both grounded strictly in on-chain evidence. It then presents MATEX, a multi-agent framework with a shared Evidence Board and four specialized agents (Profiler, Protocol Investigator, Synthesizer, Safety Auditor) to generate attributed, verifiable narratives and validate them adversarially. Across evaluations, MATEX outperforms baselines in accuracy, faithfulness, and risk identification, underscoring its practical potential for safer, clearer DeFi interactions.

Abstract

Understanding a complicated Ethereum transaction remains challenging: multi-hop token flows, nested contract calls, and opaque execution paths routinely lead users to blind signing. Based on interviews with everyday users, developers, and auditors, we identify the need for faithful, step-wise explanations grounded in both on-chain evidence and real-world protocol semantics. To meet this need, we introduce (matex, a cognitive multi-agent framework that models transaction understanding as a collaborative investigation-combining rapid hypothesis generation, dynamic off-chain knowledge retrieval, evidence-aware synthesis, and adversarial validation to produce faithful explanations.

Paper Structure

This paper contains 17 sections, 2 figures.

Table of Contents

  1. Introduction
  2. Background & Motivation
  3. Understanding Transaction Flow
  4. Case Study on Web3 Infra
  5. User Study
  6. Identified Attributes
  7. TxExplain Dataset
  8. Multi-Agent Framework: $\textit{MATEX}$
  9. Agent Design Overview
  10. The Profiler
  11. The Protocol Investigator
  12. The Synthesizer
  13. The Safety Auditor
  14. Evaluations
  15. RQs
  16. ...and 2 more sections

Figures (2)

  • Figure 1: An example of the complexity aggregator splits the trade across multiple protocols and tokens.
  • Figure 2: The $\textit{MATEX}$ cognitive architecture. Agents collaborate within a shared long-context "Evidence Board": the Profiler flags uncertainty, the Protocol Investigator retrieves off-chain knowledge (e.g., docs, verified code), the Synthesizer fuses evidence into an attributed explanation, and the Safety Auditor enables iterative refinement via simulated user critique.