Single-cell RNA sequencing (scRNA-seq) allows cell classification using genome-wide transcriptional state; however, high-dimensional transcriptomic profiles, and the unsupervised analyses employed to interpret them, provide a systematically different view of biology than well-established functional/lineage definitions of immunocytes. Understanding these differences and limits is essential for accurate interpretation of these rich data. We present the Rhesus Immune Reference Atlas (RIRA), the first immune-focused macaque single-cell multi-tissue atlas. We contrasted transcriptional profiles against immune lineages, using surface protein and marker genes as ground truth. While the pattern of clustering can align with cell type, this is not always true. Especially within T and natural killer (NK) cells, many functionally distinct subsets lack defining markers, and strong shared expression programs, such as cytotoxicity, result in systematic intermingling by unsupervised clustering. We identified gene programs with high discriminatory/diagnostic value, including multi-gene signatures that model T/NK cell maturation. Directly measuring these diagnostic programs complements unsupervised analyses.