Be more generous with convergence order in the case of gmsh-based refinement