The results show that chemical CoT is neither a faithful explanation nor merely a post-hoc rationalization, but a hallucination-prone molecular scratchpad, which cautions against treating CoT as direct evidence of faithful reasoning and motivates process-level supervision beyond answer-only evaluation.
Abstract
Chemical reasoning language models are expected to derive molecular answers through faithful chain-of-thought (CoT). However, across four reasoning model families and twelve chemistry tasks, hallucination is widespread and largely decoupled from answer correctness: correct answers often coexist with fabricated structural claims absent from the relevant molecules. Yet this does not make the reasoning trace computationally irrelevant. Attribution analyses suggest a shared scratchpad function expressed in model-specific forms: Chem-R and ether-0 rely on fragmented SMILES drafts, whereas ChemDFM-R emphasizes scaffold, positional, and naming cues. Notably, perturbing Chem-R's SMILES sketches degrades generation, showing that structural drafts can be causally load-bearing even when verbal structural claims are largely inert. Together, these results show that chemical CoT is neither a faithful explanation nor merely a post-hoc rationalization, but a hallucination-prone molecular scratchpad. This finding cautions against treating CoT as direct evidence of faithful reasoning and motivates process-level supervision beyond answer-only evaluation.
A novel, large-scale reasoning dataset of reaction mechanisms, and the FukuyamaBench, a difficult benchmark derived from Fukuyama's Advanced Organic Reaction Mechanism book, to rigorously evaluate model performance on hierarchical mechanism reasoning, demonstrate that mechanism-aware training substantially enhances chemical reasoning in language models.
Xingyu Dang, Haocheng Tang, Junmei Wang et al.· arXiv.org· 0 citations
HALO (Hallucination-Aware Layered Oversight) is presented, an assurance architecture which treats hallucination as a containable failure mode rather than an eliminable one and detail each layer, give particular attention to evidence-based confidence (which verifies extractions against the source document rather than trusting the model's self-reported certainty).
Bogdan Raduta, Horia Velicu, Alexandru Preda et al.· 0 citations
A concise two-axis framework that integrates an “intrinsic-extrinsic” distinction in source attribution introduced by Ji et al. with a “faithfulness-factuality” distinction in contextual grounding surveyed is presented, yielding four clearly defined hallucination types applicable across tasks, modalities and architectures.
Misbah Khan, Preston Billion-Polak, T. Khoshgoftaar· IEEE Access· 0 citations
A unified symbolic, behavioral, and mechanistic framework that connects symbolic triggers with internal failure dynamics in transformer architectures and provides an interpretable basis for diagnosing and stabilizing symbolic reasoning in LLMs is introduced.
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.
S. A. Hebbar, Peiyao Sheng, Sewoong Oh et al.· 0 citations
