Towards more realistic forecasting of dark energy constraints from galaxy redshift surveys