The applicability of several Reynolds averaged Navier–Stokes (RANS) turbulence models in calculating the transient evolution of a buoyancy-induced flow reversal along a vertical heated plate is analyzed through the use of validation quality experimental data from the Rotatable Buoyancy Tunnel (RoBuT) facility. This benchmark attempts to capture the transient evolution from downward forced convection to upward natural convection by removing power to the blower and allowing the buoyancy force emanating from the heated plate to gradually dominate as the primary driving force. Boundary conditions and system response quantities for the numerical model are supplied from the experiment every 0.2 s during the 18.2 s transient. ASME standards are used to quantify the numerical uncertainties while the input uncertainties are handled using a Latin hypercube sampling (LHS) method based on the steady-state conditions . Qualitative comparisons between numerical and experimental results at several downstream locations are supported using a validation metric based on the statistical disparity between the respective empirical and cumulative distribution functions (CDFs). The results from this study show that the standard linear eddy-viscosity models have difficulty in reproducing the complex features of the flow reversal in comparison with the more intricate turbulence models such as Reynolds stress models (RSM) and low-Reynolds number variants. This study also briefly highlights the difficulties of capturing validation quality data for three-dimensional multiphysics flow, while also providing insight for the design of future experimental efforts.