Same configuration as Fig. D.1. This time however, we show the drift between the Th-Ar spectra with their own wavelength solution with respect to Th-Ar master spectra. We note that we have a master Th-Ar spectrum for each of the 5 lamps used over the 15 years of HARPS, plus an additional one to compensate for the systematics induced by the change of the HARPS fibres. In this plot, we corrected for the offset that exists between each consecutive Th-Ar master spectra. The drift observed over time is due to a slow variation of the focus of the instrument, which changes the PSF and therefore the shape of the spectral lines over time in the Th-Ar spectra recorded on the CCD. The master Th-Ar spectra is by construction an average of all the Th-Ar spectra, and therefore do not account for the change in the focus of HARPS.