Test-Time Agent Evolution for Long-Horizon Legal Reasoning

2026-10-06

AI Summary

This study addresses the reliability limitations of static agent strategies in legal reasoning, which struggle with case heterogeneity and cross-role, cross-stage dependencies. We propose a training-free test-time agent evolution method that introduces a test-time memory evolution mechanism to reuse historical experiences and dynamically adapt to case specifics, while enabling multi-role collaborative decision-making through scoring-criteria-aligned cooperation. Furthermore, the approach integrates retrieval-augmented generation, experience consolidation, and verification-revision algorithms, allowing adaptive deployment signal utilization without updating model parameters. Evaluated on the J1-EVAL and LegalWorld benchmarks, our method achieves consistent performance improvements over mainstream baselines. At a reasonable computational cost, it significantly enhances the reliability and efficiency of long-horizon legal reasoning.

Abstract

Legal intelligence aims to support reliable decision-making across long-horizon legal processes involving evolving case states and multiple roles. However, real-world legal deployment exhibits substantial case heterogeneity in facts, evidence, and procedural contexts, exposing the limitations of static agent strategies. Moreover, legal reasoning is inherently interdependent across roles and procedural stages, making global reliability fundamentally different from isolated role competence. To address these challenges, we study training-free test-time agent adaptation, where agents continuously exploit deployment-time signals from preceding cases and ongoing interactions without updating model parameters. We propose \method, which introduces \emph{Test-Time Memory Evolution} to retrieve reusable experience from previous cases, adapt it to the current factual and procedural context, and consolidate accumulated experience for subsequent decision-making. Further, \emph{Rubric-Aligned Collaboration} verifies and revises role-specific actions according to behavioral and procedural requirements, enabling coordinated decision-making across roles and stages. Extensive experiments on J1-EVAL and LegalWorld across five backbone models demonstrate consistent improvements over representative reasoning and agent baselines with reasonable interaction and computational costs. Ablation and case studies further show that the two components provide complementary benefits in experience adaptation and cross-role coordination, improving the reliability and efficiency of long-horizon legal reasoning.

Problem

• long-horizon legal reasoning
• test-time adaptation
• case heterogeneity
• multi-role coordination
• training-free

Innovation

• Test-Time Agent Adaptation
• Legal Reasoning
• Memory Evolution
• Rubric-Aligned Collaboration
• Training-Free