Hallucinations on the Board: Tool-Augmented Evaluation of LLM Chess Commentary

2026-08-04

AI Summary

This work addresses the prevalence of factual hallucinations in chess commentary generated by large language models (LLMs), a challenge inadequately tackled by existing evaluation methods. To this end, the authors propose ACT-Eval, a novel tool-augmented atomic evaluation framework that decomposes commentary into atomic statements and assesses them along three dimensions: factual correctness, conceptual coverage, and move quality. The framework integrates chess engine verification, expert-annotated gold-standard labels, and an LLM-as-a-judge mechanism. A benchmark constructed under this paradigm encompasses five error categories and expert-validated atomic claims. Experiments reveal hallucination rates of 22%–40% across mainstream models, with tool augmentation substantially improving factual accuracy, though strategic and tactical coverage remains limited. Notably, ACT-Eval’s assessments exhibit strong alignment with human judgments.

Abstract

Superhuman game engines in domains like chess have made expert-level evaluations easily accessible, yet they communicate what is true without the natural-language explanations that make such expertise educationally useful to experts and non-experts alike. Large language models could, in principle, bridge this gap, but they frequently hallucinate due to limited domain-specific knowledge, and standard reference-based or LLM-as-a-judge frameworks cannot reliably detect these errors. In this work, we present ACT-Eval, an evaluation framework that decomposes chess commentary into atomic claims and routes them to engine-supported tools and expert-annotated gold references to assess factual correctness, conceptual coverage, and move-quality judgment. We release a benchmark of 325 position--move pairs spanning pedagogical, tournament, and critical positions, including 125 positions with expert-verified gold atoms and a five-class error taxonomy. Evaluating leading proprietary and open-weight models, we find that factual hallucinations remain pervasive in chess commentary: GPT-5.4 without tools produces incorrect sub-claims 22.0% of the time, while smaller open-weight models exceed 40%. Although tool augmentation substantially improves factual correctness and move-quality assessment, coverage of expert strategic and tactical ideas remains limited across all models. Human calibration shows that ACT-Eval's factual judgments fall within the observed range of inter-human agreement, while its coverage scores correlate strongly with human assessments of strategic completeness.

Problem

• hallucination
• chess commentary
• factual correctness
• evaluation framework
• large language models

Innovation

• tool-augmented evaluation
• atomic claim decomposition
• factual hallucination detection
• chess commentary benchmark
• engine-supported verification