The investigation of upper mantle structure beneath the US has revealed a growing diversity of discontinuities within, across, and underneath the sub-continental lithosphere. As the complexity and variability of these detected discontinuities increase - e.g., velocity increase/decrease, number of layers and depth - it is hard to judge which constraints are robust and which explanatory models generalize to the largest set of constraints. Much work has been done to image discontinuities of interest using S-waves that convert to P-waves (or reflect back as S-waves). A higher resolution method using P-to-S scattered waves is preferred but often obscured by multiply reflected waves trapped in a shallow layer, limiting the visibility of deeper boundaries. Here, we address the interference problem and re-evaluate upper mantle stratification using filtered Ps-RFs interpreted using unsupervised machine-learning. Robust insight into upper mantle layering is facilitated with CRISP-RF: Clean Receiver-Function Imaging using Sparse Radon Filters. Subsequent sequencing and clustering of the polarity-filtered Ps-RFs into distinct depth-based clusters, clearly distinguishes three discontinuity types: (1) intra-lithosphere discontinuity with no base, (2) intra-lithosphere discontinuity with a top and bottom boundary (3) transitional and sub-lithosphere discontinuities. Our findings contribute a more nuanced understanding of mantle discontinuities, offering new perspectives on the nature of upper mantle layering beneath continents.