Case-Based Calibration of Adaptive Reasoning and Execution for LLM Tool Use

TL;DR

CAST is a case-based framework that enhances large language models' tool use by leveraging historical execution data to improve accuracy, efficiency, and structural validity.

Contribution

It introduces a novel case-driven approach that extracts signals from past executions to optimize reasoning strategies and structural reliability during reinforcement learning.

Findings

Improves execution accuracy by up to 5.85 percentage points.

Reduces average reasoning length by 26%.

Enhances structural validity and task success in tool use.

Abstract

Tool use extends large language models beyond parametric knowledge, but reliable execution requires balancing appropriate reasoning depth with strict structural validity. We approach this problem from a case-based perspective to present CAST, a case-driven framework that treats historical execution trajectories as structured cases. Instead of reusing raw exemplar outputs, CAST extracts case-derived signals to identify complexity profiles for estimating optimal reasoning strategies, alongside failure profiles to map likely structural breakdowns. The framework translates this knowledge into a fine-grained reward design and adaptive reasoning, enabling the model to autonomously internalize case-based strategies during reinforcement learning. Experiments on BFCLv2 and ToolBench demonstrate that CAST improves both schema-faithful execution and task-level tool-use success while reducing unnecessary deliberation. The approach achieves up to 5.85 percentage points gain in overall execution accuracy and reduces average reasoning length by 26%, significantly mitigating high-impact structural errors. Ultimately, this demonstrates how historical execution cases can provide reusable adaptation knowledge for calibrated tool use.