On Dimensional Analyses of Bone Surface Modifications, Machine Learning, and Straw Men

The identification and interpretation of bone surface modifications (BSM) are central to reconstructing early hominin behavior, yet recent shifts by some researchers toward metric quantification face significant epistemological and statistical challenges. This paper critically evaluates the “metric method” proposed by Keevil/Pante et al., arguing that its reliance on continuous, ratio-scale measurements is fundamentally undermined by effector variance—the inherent dimensional mismatch between experimental tools and those in the (assemblage-specific) archaeological record. I demonstrate through statistical analysis that the method’s use of quadratic discriminant analysis (QDA) is compromised by severe multicollinearity (VIF > 5), resulting in unstable models that fail to generalize to fossil contexts, as exemplified by the problematic interpretations of the Grăunceanu (Romania) assemblage. Furthermore, I deconstruct recent critiques against the application of machine learning (ML) in taphonomy, clarifying misconceptions regarding Wolpert’s “no free lunch” theorem and rebutting allegations of data leakage. I contend that ML algorithms, when properly integrated with high-resolution categorical data, provide a more robust and accurate framework for classification of taphonomic datasets. By exposing these methodological biases, I propose a new paradigm for BSM analysis that prioritizes systemic hypothesis testing, objective data generation, and empirical testing/refutation models based on original datasets over speculative arguments and strict metric quantification.