A quantity of interest changes substantially when a CFD mesh is refined once. What is the most defensible next step before treating its predicted value as mesh-independent?