Substitution model selection is central to phylogenetic inference and is commonly treated as a problem of identifying the substitutional complexity required to describe sequence evolution along a single tree. This framework implicitly assumes a shared genealogy across all sites, an assumption that is routinely violated in phylogenomic data by incomplete lineage sorting and other sources of gene-tree discordance. Recent work by Lozano et al. (2026) demonstrates that unmodeled genealogical heterogeneity can systematically distort substitution model selection, creating spurious support for parameter-rich models even when the underlying substitution process is simple. In this perspective, we examine the consequences of this confounding for phylogenetic estimation, divergence-time inference, and the biological interpretation of substitution model parameters, highlighting the need to more explicitly integrate genealogical heterogeneity into model selection and adequacy assessment in phylogenomics. This perspective reframes substitution model choice not only as a tool for describing molecular evolution, but also as a potential diagnostic for violations of shared-genealogy assumptions that are central to modern systematics.
