The authorship of the Pauline epistles, particularly the Pastoral Epistles (1 and 2 Timothy and Titus), remains contested in biblical scholarship. Quantitative stylometry can contribute to this debate, but supervised approaches risk circular reasoning when they depend on predefined "core Pauline" labels, and lexical measures risk topic bias when distinctive vocabulary reflects subject matter rather than authorial style. This study proposes a transparent, sensitivity-oriented workflow for examining stylistic boundaries within the New Testament corpus. Using lemmatized Greek texts, it combines BM25 weighting, principal component analysis, hierarchical and k-means clustering, sensitivity analysis across alternative definitions of the Core Pauline letters, function-word-only analysis, and SHAP-based model interpretation. The results show that the Pastoral Epistles consistently occupy a region of stylometric space distinct from the Core Pauline letters; Deutero-Pauline letters tend to occupy a more intermediate position, while the Pastorals remain in the lowest tier of relative Core-likeness across core definitions. The pattern persists when content vocabulary is removed, suggesting that the observed divergence is not solely attributable to a topic-specific lexicon. The study does not claim to determine historical authorship definitively, especially given the absence of an externally verified Pauline stylome. Instead, it provides convergent quantitative evidence for a stable stylistic boundary that any historical account of the Pastorals—single-author, amanuensis-mediated, Pauline-school, or pseudonymous—must explain. The workflow illustrates how explainable machine learning can support transparent and reproducible stylometric analysis in small historical corpora. An earlier extended version is available at https://doi.org/10.5281/zenodo.18503648.
Katsuro ANAZAWA (Fri,) studied this question.