The surge in the uptake of harmonic producing loads in 21st century smart distribution networks has necessitated that robust data-driven approaches for harmonic assessment be developed. Moreover, there exists abundant harmonic data which can be leveraged on for the construction of this harmonic assessment tool. This is the objective of this paper. Large volumes of time-stamped data acquired from a practical distribution network in Edmonton, Canada, have been used to construct a time-dependent cross-coupled harmonic model that has been combined with a formulated iterative time-dependent robust extended computer-aided harmonic power flows. The constructed harmonic power flow formulation also considers the constraints relating to the power conservation principle at the fundamental frequency. Practical network conditions including untransposed lines, load unbalance and skin effects of conductors have been thoroughly modelled. The proposed method has been applied to practical radial and weakly-meshed medium voltage distribution networks and it demonstrated robustness to initialisation of the iterative procedure. Moreover, findings from this detailed data-driven technique reveal that harmonic impacts in the distribution networks depend on the quantum of harmonic levels, type of network and location of the harmonic sources.