# L-MARS: Legal Multi-Agent System with Agentic Search and Citation-Faithfulness Audit

DevFeed: [L-MARS: Legal Multi-Agent System with Agentic Search and Citation-Faithfulness Audit](<https://devfeed.tech/articles/l-mars-legal-multi-agent-system-with-agentic-search-and-citation-faithfulness-audit-124353.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2509.00761>)

Author: Boqin Yuan, Ziqi Wang

Published: 2026-11-09T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Multi Agent Systems](<https://devfeed.tech/topics/multi-agent-systems.md>), [audit](<https://devfeed.tech/topics/audit.md>), [Agentic AI](<https://devfeed.tech/topics/what-is-agentic-ai.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [atomic](<https://devfeed.tech/tags/atomic.md>), [audit](<https://devfeed.tech/tags/audit.md>), [case-study](<https://devfeed.tech/tags/case-study.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [evaluations](<https://devfeed.tech/tags/evaluations.md>), [exam](<https://devfeed.tech/tags/exam.md>), [language](<https://devfeed.tech/tags/language.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [legal](<https://devfeed.tech/tags/legal.md>), [model](<https://devfeed.tech/tags/model.md>), [multi-agent](<https://devfeed.tech/tags/multi-agent.md>), [multi-agent-system](<https://devfeed.tech/tags/multi-agent-system.md>), [rag](<https://devfeed.tech/tags/rag.md>), [search](<https://devfeed.tech/tags/search.md>)

## AI overview

L-MARS audits legal answers claim by claim against their cited sources, using a multi-agent search system and cross-provider judging. In a stratified 100-question Bar Exam audit, a multi-turn judge loop raised strict citation F1 from 0.13 for naive RAG to 0.25 and reduced answers without citations from 34% to 13%, while retrieval alone barely changed accuracy. Its Faith-Search post-draft step repaired unreachable citations, lowering their rate below 1%, but did not improve citation F1 over the judge loop. In a 50-question LegalSearchQA case study, retrieve-then-draft systems reached about 0.75 citation F1, while a single-agent web-search baseline scored 0.22 under external audit. The abstract presents Faith-Search as a reachability fix, not evidence of improved citation support.

## Source excerpt

arXiv:2509.00761v5 Announce Type: replace Abstract: Large language models are increasingly deployed for legal question answering, where evaluations typically focus on multiple-choice accuracy. This measure overlooks a common failure: whether the citation source attached to an answer exists and supports the rule the system attributes to it. We present L-MARS, an open multi-agent legal QA system with agentic search and judge-driven evidence checks, and audit it claim by claim against its cited source. Each atomic claim is labelled with a six-class taxonomy and scored with strict-ALCE under cross-provider judging, where the answerer and verifier come from different model families. On a stratified 100-question Bar Exam audit, retrieval barely moves accuracy, yet the multi-turn judge loop lifts strict citation F1 from 0.13 (naive RAG) to 0.25 and cuts the no-citation rate from 34% to 13%. We further introduce Faith-Search, a post-draft step that re-verifies and repairs unreachable citations; it drops the unreachable rate below 1% but does not improve F1 over the multi-turn loop, so we report it as a targeted reachability intervention rather than a faithfulness breakthrough. A 50-question LegalSearchQA case study confirms the picture: retrieve-then-draft pipelines saturate near 0.75 citation F1, while a single-agent web-search baseline collapses to 0.22 under external audit.