Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

L-MARS: Legal Multi-Agent Workflow with Orchestrated Reasoning and Agentic Search

Ziqi Wang, Boqin Yuan

Abstract

We present L-MARS (Legal Multi-Agent Workflow with Orchestrated Reasoning and Agentic Search), a multi-agent retrieval framework for grounded legal question answering that decomposes queries into structured sub-problems, retrieves evidence via agentic web search, filters results through a verification agent, and synthesizes cited answers. Existing legal QA benchmarks test either closed-book reasoning or retrieval over fixed corpora, but neither captures scenarios requiring current legal information. We introduce LegalSearchQA, a 50-question benchmark across five legal domains whose answers depend on recent developments that post-date model training data. L-MARS achieves 96.0% accuracy on LegalSearchQA, a 38.0% improvement over zero-shot performance (58.0%), while chain-of-thought prompting degrades performance to 30.0%. On Bar Exam QA (Zheng et al., 2025), a reasoning-focused benchmark of 594 bar examination questions, retrieval provides negligible gains (+0.7 percentage points), consistent with prior findings. These results show that agentic retrieval dramatically improves legal QA when tasks require up-to-date factual knowledge, but the benefit is benchmark-dependent, underscoring the need for retrieval-focused evaluation. Code and data are available at: https://github.com/boqiny/L-MARS

L-MARS: Legal Multi-Agent Workflow with Orchestrated Reasoning and Agentic Search

Abstract

We present L-MARS (Legal Multi-Agent Workflow with Orchestrated Reasoning and Agentic Search), a multi-agent retrieval framework for grounded legal question answering that decomposes queries into structured sub-problems, retrieves evidence via agentic web search, filters results through a verification agent, and synthesizes cited answers. Existing legal QA benchmarks test either closed-book reasoning or retrieval over fixed corpora, but neither captures scenarios requiring current legal information. We introduce LegalSearchQA, a 50-question benchmark across five legal domains whose answers depend on recent developments that post-date model training data. L-MARS achieves 96.0% accuracy on LegalSearchQA, a 38.0% improvement over zero-shot performance (58.0%), while chain-of-thought prompting degrades performance to 30.0%. On Bar Exam QA (Zheng et al., 2025), a reasoning-focused benchmark of 594 bar examination questions, retrieval provides negligible gains (+0.7 percentage points), consistent with prior findings. These results show that agentic retrieval dramatically improves legal QA when tasks require up-to-date factual knowledge, but the benefit is benchmark-dependent, underscoring the need for retrieval-focused evaluation. Code and data are available at: https://github.com/boqiny/L-MARS

Paper Structure

This paper contains 36 sections, 1 equation, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Agentic RAG and Search Systems
  4. Legal AI and Benchmarks
  5. Methodology
  6. Problem Formation
  7. Workflow and Agents Overview
  8. Workflow
  9. Agents
  10. Agentic Search
  11. Online Search
  12. Local RAG Indexing
  13. CourtListener Integration
  14. L-MARS Operating Mode
  15. Simple Mode
  16. ...and 21 more sections

Figures (3)

  • Figure 1: Comparison of the two L-MARS operating modes. (a) Simple Mode: executes a single-pass retrieval–summarization pipeline. (b) Multi-Turn Mode: adds Judge Agent–guided iterations with sufficiency checks and query refinement.
  • Figure 2: Per-domain accuracy on LegalSearchQA. L-MARS achieves near-perfect accuracy across all domains. CoT falls below the 25% random baseline in three of five domains.
  • Figure 3: Benchmark comparison. On retrieval-dependent LegalSearchQA, L-MARS gains 38 percentage points over zero-shot. On reasoning-focused Bar Exam QA, retrieval provides negligible improvement.