Simulation-based nuclear reactor design requires highly efficient codes that quantify the requisite physics while having the efficiency required for optimization-based design and uncertainty quantification. To achieve the required accuracy and predictive capabilities, phenomenological parameters, often employed in closure relations or to quantify unmodeled or unresolved physics, must be calibrated for considered reactor conditions and designs. When available, experimental data with quantified observation errors are ideally employed for calibration. However, for many thermal-hydraulic, fuel, and Chalk River Unidentified Deposits modeling regimes, experimental data are prohibitively expensive or impossible to collect. For such cases, we demonstrate the use of a mutual information–based experimental design framework to employ validated high-fidelity codes to calibrate parameters in low-fidelity design codes. We demonstrate the use of the high-fidelity computational fluid dynamics package STAR-CCM+ to calibrate the turbulent mixing coefficient β in COBRA-TF (CTF). This includes the construction and verification of a surrogate for CTF, which permits the computationally intensive experimental design and Bayesian calibration steps. We also demonstrate Bayesian inference of parameter distributions for the Dittus-Boelter relation and propagation of these uncertainties through CTF to improve uncertainty bounds for computed maximum fuel temperatures.