EG-MAP: Evidence-Grounded Multi-Agent Smart Contract Auditing

Zhuoli SONG; Linhe YUE; Zejia TAO; Qiyun HAN; Wenjie ZENG; Zhiyu XU; Shantanu PAL; Chengzu DONG

doi:10.1109/SSE72781.2026.00018

EG-MAP: Evidence-Grounded Multi-Agent Smart Contract Auditing

Zhuoli SONG
, Linhe YUE
, Zejia TAO
, Qiyun HAN
, Wenjie ZENG
, Zhiyu XU^*
, Shantanu PAL
, Chengzu DONG

^*Corresponding author for this work

School of Data Science

Research output: Book Chapters | Papers in Conference Proceedings › Conference paper (refereed) › Referred Conference Paper › peer-review

Abstract

Smart contracts are core infrastructure for decentralized software services, especially in decentralized finance (DeFi), where vulnerabilities can cause severe financial losses. Detecting semantic flaws in these contracts remains difficult because many failures arise from protocol logic rather than local syntax. Large language models (LLMs) can reason about program intent, but unguided use often produces speculative findings and hallucination-driven false positives. Here, we propose EG-MAP, an evidence-grounded multi-agent platform for smart contract auditing. EG-MAP uses Planner, Analyst, and Verifier service nodes to decompose audit scope, generate vulnerability hypotheses, and validate evidence. To support auditable service decisions, we introduce a computable evidence-gating lattice with Dempster-Shafer conflict fusion. This mechanism integrates LLM reasoning, static-analysis hints, and verifier evidence before a finding is accepted. EG-MAP records structured evidence traces, including hypotheses, verifier assessments, lattice states, and belief scores. We evaluate EG-MAP on two public benchmark suites and five real-world exploit cases. Under fuzzy matching, EG-MAP achieves F1-scores of 0.738 on SmartBugs-Curated and 0.469 on SolidiFI.

Original language	English
Title of host publication	2026 IEEE International Conference on Software Services Engineering : Proceedings
Editors	Rong N. CHANG, Carl K. CHANG, Michael SHENG, Xiaofei XU, Boualem BENATALLAH, Zhongjie WANG, Nimanthi ATUKORALA, Mingyi LIU, Sira YONGCHAREON, Tiberiu SECELEANU, Jun SHEN, Shangguang WANG, Xiao LIU, Yingjie WANG, Sami YANGUI, Shijun LIU, Stephan REIFF-MARGANIEC
Publisher	IEEE
Pages	78-88
Number of pages	11
ISBN (Electronic)	9798319511980
DOIs	https://doi.org/10.1109/SSE72781.2026.00018
Publication status	Published - Jul 2026
Event	2026 IEEE International Conference on Software Services Engineering - Sydney, Australia Duration: 13 Jul 2026 → 18 Jul 2026

Conference

Conference	2026 IEEE International Conference on Software Services Engineering
Abbreviated title	IEEE SSE 2026
Country/Territory	Australia
City	Sydney
Period	13/07/26 → 18/07/26

Bibliographical note

Publisher Copyright:
©2026 IEEE.

Keywords

Smart Contract Auditing Service
Logic Vulnerabilities
Multi-Agent Service
Dempster-Shafer Theory
Automated Program Analysis

Access to Document

10.1109/SSE72781.2026.00018

Cite this

SONG, Z., YUE, L., TAO, Z., HAN, Q., ZENG, W., XU, Z., PAL, S., & DONG, C. (2026). EG-MAP: Evidence-Grounded Multi-Agent Smart Contract Auditing. In R. N. CHANG, C. K. CHANG, M. SHENG, X. XU, B. BENATALLAH, Z. WANG, N. ATUKORALA, M. LIU, S. YONGCHAREON, T. SECELEANU, J. SHEN, S. WANG, X. LIU, Y. WANG, S. YANGUI, S. LIU, & S. REIFF-MARGANIEC (Eds.), 2026 IEEE International Conference on Software Services Engineering : Proceedings (pp. 78-88). IEEE. https://doi.org/10.1109/SSE72781.2026.00018

SONG, Zhuoli ; YUE, Linhe ; TAO, Zejia et al. / EG-MAP: Evidence-Grounded Multi-Agent Smart Contract Auditing. 2026 IEEE International Conference on Software Services Engineering : Proceedings. editor / Rong N. CHANG ; Carl K. CHANG ; Michael SHENG ; Xiaofei XU ; Boualem BENATALLAH ; Zhongjie WANG ; Nimanthi ATUKORALA ; Mingyi LIU ; Sira YONGCHAREON ; Tiberiu SECELEANU ; Jun SHEN ; Shangguang WANG ; Xiao LIU ; Yingjie WANG ; Sami YANGUI ; Shijun LIU ; Stephan REIFF-MARGANIEC. IEEE, 2026. pp. 78-88

@inproceedings{63bd79ba73594281b78bc5285db4fcfb,

title = "EG-MAP: Evidence-Grounded Multi-Agent Smart Contract Auditing",

abstract = "Smart contracts are core infrastructure for decentralized software services, especially in decentralized finance (DeFi), where vulnerabilities can cause severe financial losses. Detecting semantic flaws in these contracts remains difficult because many failures arise from protocol logic rather than local syntax. Large language models (LLMs) can reason about program intent, but unguided use often produces speculative findings and hallucination-driven false positives. Here, we propose EG-MAP, an evidence-grounded multi-agent platform for smart contract auditing. EG-MAP uses Planner, Analyst, and Verifier service nodes to decompose audit scope, generate vulnerability hypotheses, and validate evidence. To support auditable service decisions, we introduce a computable evidence-gating lattice with Dempster-Shafer conflict fusion. This mechanism integrates LLM reasoning, static-analysis hints, and verifier evidence before a finding is accepted. EG-MAP records structured evidence traces, including hypotheses, verifier assessments, lattice states, and belief scores. We evaluate EG-MAP on two public benchmark suites and five real-world exploit cases. Under fuzzy matching, EG-MAP achieves F1-scores of 0.738 on SmartBugs-Curated and 0.469 on SolidiFI.",

keywords = "Smart Contract Auditing Service, Logic Vulnerabilities, Multi-Agent Service, Dempster-Shafer Theory, Automated Program Analysis",

author = "Zhuoli SONG and Linhe YUE and Zejia TAO and Qiyun HAN and Wenjie ZENG and Zhiyu XU and Shantanu PAL and Chengzu DONG",

note = "Publisher Copyright: {\textcopyright}2026 IEEE.; 2026 IEEE International Conference on Software Services Engineering, IEEE SSE 2026 ; Conference date: 13-07-2026 Through 18-07-2026",

year = "2026",

month = jul,

doi = "10.1109/SSE72781.2026.00018",

language = "English",

pages = "78--88",

editor = "CHANG, \{Rong N.\} and CHANG, \{Carl K.\} and Michael SHENG and Xiaofei XU and Boualem BENATALLAH and Zhongjie WANG and Nimanthi ATUKORALA and Mingyi LIU and Sira YONGCHAREON and Tiberiu SECELEANU and Jun SHEN and Shangguang WANG and Xiao LIU and Yingjie WANG and Sami YANGUI and Shijun LIU and Stephan REIFF-MARGANIEC",

booktitle = "2026 IEEE International Conference on Software Services Engineering : Proceedings",

publisher = "IEEE",

}

SONG, Z, YUE, L, TAO, Z, HAN, Q, ZENG, W, XU, Z, PAL, S & DONG, C 2026, EG-MAP: Evidence-Grounded Multi-Agent Smart Contract Auditing. in RN CHANG, CK CHANG, M SHENG, X XU, B BENATALLAH, Z WANG, N ATUKORALA, M LIU, S YONGCHAREON, T SECELEANU, J SHEN, S WANG, X LIU, Y WANG, S YANGUI, S LIU & S REIFF-MARGANIEC (eds), 2026 IEEE International Conference on Software Services Engineering : Proceedings. IEEE, pp. 78-88, 2026 IEEE International Conference on Software Services Engineering, Sydney, New South Wales, Australia, 13/07/26. https://doi.org/10.1109/SSE72781.2026.00018

EG-MAP: Evidence-Grounded Multi-Agent Smart Contract Auditing. / SONG, Zhuoli; YUE, Linhe; TAO, Zejia et al.
2026 IEEE International Conference on Software Services Engineering : Proceedings. ed. / Rong N. CHANG; Carl K. CHANG; Michael SHENG; Xiaofei XU; Boualem BENATALLAH; Zhongjie WANG; Nimanthi ATUKORALA; Mingyi LIU; Sira YONGCHAREON; Tiberiu SECELEANU; Jun SHEN; Shangguang WANG; Xiao LIU; Yingjie WANG; Sami YANGUI; Shijun LIU; Stephan REIFF-MARGANIEC. IEEE, 2026. p. 78-88.

Research output: Book Chapters | Papers in Conference Proceedings › Conference paper (refereed) › Referred Conference Paper › peer-review

TY - GEN

T1 - EG-MAP: Evidence-Grounded Multi-Agent Smart Contract Auditing

AU - SONG, Zhuoli

AU - YUE, Linhe

AU - TAO, Zejia

AU - HAN, Qiyun

AU - ZENG, Wenjie

AU - XU, Zhiyu

AU - PAL, Shantanu

AU - DONG, Chengzu

PY - 2026/7

Y1 - 2026/7

N2 - Smart contracts are core infrastructure for decentralized software services, especially in decentralized finance (DeFi), where vulnerabilities can cause severe financial losses. Detecting semantic flaws in these contracts remains difficult because many failures arise from protocol logic rather than local syntax. Large language models (LLMs) can reason about program intent, but unguided use often produces speculative findings and hallucination-driven false positives. Here, we propose EG-MAP, an evidence-grounded multi-agent platform for smart contract auditing. EG-MAP uses Planner, Analyst, and Verifier service nodes to decompose audit scope, generate vulnerability hypotheses, and validate evidence. To support auditable service decisions, we introduce a computable evidence-gating lattice with Dempster-Shafer conflict fusion. This mechanism integrates LLM reasoning, static-analysis hints, and verifier evidence before a finding is accepted. EG-MAP records structured evidence traces, including hypotheses, verifier assessments, lattice states, and belief scores. We evaluate EG-MAP on two public benchmark suites and five real-world exploit cases. Under fuzzy matching, EG-MAP achieves F1-scores of 0.738 on SmartBugs-Curated and 0.469 on SolidiFI.

AB - Smart contracts are core infrastructure for decentralized software services, especially in decentralized finance (DeFi), where vulnerabilities can cause severe financial losses. Detecting semantic flaws in these contracts remains difficult because many failures arise from protocol logic rather than local syntax. Large language models (LLMs) can reason about program intent, but unguided use often produces speculative findings and hallucination-driven false positives. Here, we propose EG-MAP, an evidence-grounded multi-agent platform for smart contract auditing. EG-MAP uses Planner, Analyst, and Verifier service nodes to decompose audit scope, generate vulnerability hypotheses, and validate evidence. To support auditable service decisions, we introduce a computable evidence-gating lattice with Dempster-Shafer conflict fusion. This mechanism integrates LLM reasoning, static-analysis hints, and verifier evidence before a finding is accepted. EG-MAP records structured evidence traces, including hypotheses, verifier assessments, lattice states, and belief scores. We evaluate EG-MAP on two public benchmark suites and five real-world exploit cases. Under fuzzy matching, EG-MAP achieves F1-scores of 0.738 on SmartBugs-Curated and 0.469 on SolidiFI.

KW - Smart Contract Auditing Service

KW - Logic Vulnerabilities

KW - Multi-Agent Service

KW - Dempster-Shafer Theory

KW - Automated Program Analysis

UR - https://www.scopus.com/pages/publications/105050755908

U2 - 10.1109/SSE72781.2026.00018

DO - 10.1109/SSE72781.2026.00018

M3 - Conference paper (refereed)

SP - 78

EP - 88

BT - 2026 IEEE International Conference on Software Services Engineering : Proceedings

A2 - CHANG, Rong N.

A2 - CHANG, Carl K.

A2 - SHENG, Michael

A2 - XU, Xiaofei

A2 - BENATALLAH, Boualem

A2 - WANG, Zhongjie

A2 - ATUKORALA, Nimanthi

A2 - LIU, Mingyi

A2 - YONGCHAREON, Sira

A2 - SECELEANU, Tiberiu

A2 - SHEN, Jun

A2 - WANG, Shangguang

A2 - LIU, Xiao

A2 - WANG, Yingjie

A2 - YANGUI, Sami

A2 - LIU, Shijun

A2 - REIFF-MARGANIEC, Stephan

PB - IEEE

T2 - 2026 IEEE International Conference on Software Services Engineering

Y2 - 13 July 2026 through 18 July 2026

ER -

SONG Z, YUE L, TAO Z, HAN Q, ZENG W, XU Z et al. EG-MAP: Evidence-Grounded Multi-Agent Smart Contract Auditing. In CHANG RN, CHANG CK, SHENG M, XU X, BENATALLAH B, WANG Z, ATUKORALA N, LIU M, YONGCHAREON S, SECELEANU T, SHEN J, WANG S, LIU X, WANG Y, YANGUI S, LIU S, REIFF-MARGANIEC S, editors, 2026 IEEE International Conference on Software Services Engineering : Proceedings. IEEE. 2026. p. 78-88 doi: 10.1109/SSE72781.2026.00018

EG-MAP: Evidence-Grounded Multi-Agent Smart Contract Auditing

Abstract

Conference

Bibliographical note

Keywords

Access to Document

Other files and links

Fingerprint

Cite this