The reliable production of _in vitro_ chondrocytes that faithfully recapitulate _in vivo_ development would be of great benefit for orthopaedic disease modelling and regenerative therapy(1,2). Current efforts are limited by off-target differentiation, resulting in a heterogeneous product, and by the lack of comparison to human tissue, which precludes detailed evaluation of _in vitro_ cells(3,4). We performed single-cell RNA-sequencing of long bones dissected from first-trimester fetal limbs to form a detailed ‘atlas’ of endochondral ossification. Through 100-gene in-situ sequencing, we placed each sequenced cell type into its anatomical context to spatially resolve the process of endochondral ossification. We then used this atlas to perform deconvolution on a series of previously published bulk transcriptomes generated from _in vitro_ chondrogenesis protocols to evaluate their ability to accurately produce chondrocytes. We then applied single-nuclear RNA-sequencing to cells from the best performing protocol collected at multiple time points to allow direct comparison between the differentiation of _in vitro_ and _in vivo_ cells. We captured 275,000 single fetal cells, profiling the development of chondrocytes from multipotent mesenchymal progenitors to hypertrophic cells at full transcriptomic breadth. Using this atlas as the ground truth for evaluating _in vitro_ cells, we found substantial variability in cell states produced by each protocol, with many showing little similarity to _in vivo_ cells, and all exhibiting off-target differentiation. Trajectory alignment between _in vivo_ and _in vitro_ single-cell data revealed key differences in gene expression dynamics between _in vitro_ and _in vivo cells,_ with several osteoblastic transcription factors erroneously unregulated _in vitro,_ including _FOXO1._ Using this information, we inhibited _FOXO1_ in culture to successfully increase chondrocyte yield _in vitro._ This study presents a new framework for evaluating tissue engineering protocols, using single-cell data to drive improvement and bring the prospect of true engineered cartilage closer to reality.
