KDASO: Knowledge-Data Dual-Driven Automated Skill Optimization for Liability Adjudication Tasks

Authors: Zi-Jian Cheng, Weiming Wu, Jie Meng, Zhen Peng, Shan Huang, Qun Li, Guobing Wu, Lan-Zhe Guo

arXiv: 2611.06972 [cs.AI] · Submitted 4 November 2026

Subjects: Artificial Intelligence (cs.AI); Computers and Society (cs.CY)

Abstract

Liability adjudication tasks—such as legal charge determination—require analyzing event descriptions against domain-specific skills to determine whether a party should bear liability. While Large Language Models (LLMs) have shown promise in this area by leveraging executable skills for interpretable reasoning, existing approaches face three critical limitations:

• Stale skill libraries as real-world scenarios dynamically evolve;
• A semantic execution gap in skill descriptions, where LLMs deviate from intended expert logic;
• Context budget overflow as growing skill collections exceed context windows, causing critical skill omission.

To address these challenges, we propose Knowledge-Data Dual-Driven Automated Skill Optimization (KDASO), which continuously evolves a skill library through three mechanisms:

1. Neighborhood-conditioned induction from recent adjudication data;
2. Execution-aligned refinement based on LLM trajectory analysis;
3. Context-budgeted routing for scalable skill selection.



Experiments across ride-hailing adjudication, legal liability determination, and logical reasoning judgment demonstrate that KDASO significantly outperforms current general-purpose and domain-specific baselines. KDASO further achieves an 8.96% improvement when transferring skills to concept-similar scenarios and maintains consistent performance gains through continuous optimization—underscoring the high transferability and stability of the generated skills.




Key Contributions



• A dual-driven framework that jointly leverages knowledge and data to automatically optimize executable skills for LLM-based adjudication.
• Three complementary optimization mechanisms—induction, execution-aligned refinement, and budget-aware routing—that keep skill libraries current, semantically faithful to expert logic, and scalable within context limits.
• Empirical validation across three distinct domains, with strong transferability (8.96% gain on concept-similar scenarios) and stable gains under continuous optimization.



Citation




arXiv:2611.06972 [cs.AI]
https://doi.org/10.48550/arXiv.2611.06972

via ArXiv AI