<p>Fallacy identification has long been theorized as context-sensitive, stance-dependent judgment, yet computational approaches often treat it as neutral or theory-independent. This paper proposes a multi-method framework that operationalizes stance-dependence in fallacy judgments through a computationally assisted case study of Leo Tolstoy’s <i>What Is Art?</i>. Grounded in Rhetorical Structure Theory (RST), we annotate and extract elaborative argumentative sequences anchored on a single Nucleus, highlighting regions of discourse where rhetorical amplification may obscure inferential progression. These sequences are evaluated using stance-conditioned large language model (LLM) prompting, simulating eight analytically distinct reader perspectives varying in value alignment, author-related attitude, argumentation expertise, and fallacy-detection instruction. The analyses reveal systematic divergence in fallacy judgments across stances, suggesting that perceived fallacies depend not only on the text itself but also on the interaction between discourse structure and interpretive stance. To trace the basis of these judgments, we apply Toulmin’s model to identify implicit warrants, missing qualifiers, and culturally contingent backings that underlie contested evaluations. This study advances argumentation research in two ways. First, it demonstrates how computational tools can illuminate, rather than obscure, the stance-dependence of fallacy judgments. Second, it proposes a pluralistic framework that integrates rhetorical, logical, and interpretive perspectives for analyzing complex philosophical argumentation.</p>