Drift of HARPS as a function of time measured on the Th-Ar spectra taken each night to derive the wavelength solution of the detector. Each subplot represents the drift in RVs measured on the Th-Ar emission lines falling in different regions of the HARPS detector. The HARPS detector is cut into 24 regions, 16 corresponding to the blue CCD shown with a blue background (pixel 0–1024, 1024–2048, 2048–3072, 3072–4096 and orders 0–12, 12–24, 24–36 and 36–47) and 8 corresponding to the red CCD shown with a red background (same separation in pixel but for order 47–60 and 60–72). The main feature that we see here is the huge RV offset on JD = 2457174 (1st of June 2015), which is induced by the change of the HARPS fibres from circular to octagonal.