Notes.The profiles are fit by a Gaussian model using a Levenberg–Marquardt algorithm. The error bar on each cross-correlation coefficient is approximated empirically by measuring the standard deviation at systemic velocities away from the orbital velocity of the planet (200 < |v| < 1000 km s⁻¹), from which the uncertainty intervals on the best-fit parameters are derived. This empirical treatment of the standard deviation is in line with previous studies that employ the cross-correlation technique, (e.g. Hoeijmakers et al. 2015, 2018b). The second to fourth columns show the best-fit Gaussian amplitude, systemic velocity (equivalent to the planet rest-frame velocity at zero phase) and FWHM. These parameters are broadly consistent with those reported by Cauley et al. (2019), who report line depths between 0.1 and 0.5% and FWHM widths between 20 and 40 km s⁻¹ for Mg I, Ti II and Fe II. The fifth column denotes the ratio between the best-fit amplitude and the line-strength of the injected model. The sixth column provides the altitude difference between the continuum and the line centre in units of planetaryradii. The last column expresses this altitude difference as the pressure at the line peak, assuming our isothermal hydrostatic atmospheric model which predicts that the continuum is formed by absorption by H⁻ at a pressure of 3.5 mbar. The last two columns also use the planetary and stellar parameters from Gaudi et al. (2017). The uncertainty intervals on all parameters correspond to Gaussian 1σ intervals. These include the uncertainty on the fitting parameters but not the known uncertainties on the planetary and stellar parameters, nor systematic errors resulting from our choice of model.