In vitro stem cell based models of human development and disease can be illuminating only in so much as they resemble their in vivo counterparts. As such, considerable effort is taken to establish differentiation protocols that effectively derive only the cell types of interest. In particular, consideration must be given to unwanted lineages with related developmental ontologies that often emerge as sources of contamination. At end point, differentiation protocols are typically assessed through histochemical analysis of marker gene expression. But this approach limits the scope of markers that can be tested, and requires prior knowledge of the desired and contaminating cell types. We have analyzed single-cell RNA-sequencing data from 75 thousand gastrulating mouse embryos between E6.5 and E8.5, identifying transcriptional trajectories specific to each major lineage emerging from the epiblast. We then process bulk-RNA sequencing from an array of human stem cell differentiation protocols, perform similarity comparisons against the full cohort of lineage-defining gene signatures, and infer the presence of target or contaminating lineages. Using this strategy, we are able to quickly and sensitively assess the fidelity of a given stem cell protocol, transcriptome-wide and unguided, with no requirement for prior knowledge.