Post-fit timing residuals of PSR J1618−3921. Upper two plots: Post-fit timing residuals as a function of time for the best-fit timing solution (TEMPONEST fit without removing the 100 coefficient DM noise model) of PSR J1618−3921 given in Table 3. Lower plot: Post-fit timing residuals after subtracting the DM noise model. The colours denote the different backends and systems as listed in Table 2. In all plots, the uncertainties are re-scaled with the White Noise parameters EFAC and EQUAD (cf. Table 2). The middle plot also contains the time domain realisation of the 100 parameter DM noise model: the blue lines show the 100 randomly created model realisations, and the black dots indicate the median across all these at each ToA.