On the model dependence of fiducial cross-section measurements