Beyond Accuracy: A Cognitive Load Framework for Mapping the Capability Boundaries of Tool-use Agents

TL;DR

This paper introduces a cognitive load framework for evaluating large language models' tool-use capabilities, moving beyond accuracy metrics to diagnose performance bottlenecks and map model boundaries under varying task complexities.

Contribution

It presents a novel framework based on Cognitive Load Theory, including a Tool Interaction Graph and a benchmark with adjustable load, to better understand model limitations.

Findings

Models exhibit performance cliffs at high cognitive loads.

The framework's predictions align closely with empirical results.

It enables precise mapping of model capability boundaries.

Abstract

The ability of Large Language Models (LLMs) to use external tools unlocks powerful real-world interactions, making rigorous evaluation essential. However, current benchmarks primarily report final accuracy, revealing what models can do but obscuring the cognitive bottlenecks that define their true capability boundaries. To move from simple performance scoring to a diagnostic tool, we introduce a framework grounded in Cognitive Load Theory. Our framework deconstructs task complexity into two quantifiable components: Intrinsic Load, the inherent structural complexity of the solution path, formalized with a novel Tool Interaction Graph; and Extraneous Load, the difficulty arising from ambiguous task presentation. To enable controlled experiments, we construct ToolLoad-Bench, the first benchmark with parametrically adjustable cognitive load. Our evaluation reveals distinct performance cliffs as cognitive load increases, allowing us to precisely map each model's capability boundary. We validate that our framework's predictions are highly calibrated with empirical results, establishing a principled methodology for understanding an agent's limits and a practical foundation for building more efficient systems.