Top panels: Temperature difference between “ambi” and “clean” runs as a function of time. The temperature difference is averaged over heights where the percentage of points satisfying the criteria $Q_{\mathrm{AMB}}>Q_{\mathrm{filt}}^{\mathrm{VISC}}+Q_{\mathrm{hyp}}^{\mathrm{VISC}}+Q_{\mathrm{filt}}^{\mathrm{MAG}}$ is at least 20% of the maximum number of points satisfying this criteria for a given model (as per Fig. 11). The curves are smoothed over five data points to decrease the scatter and better show the trends. Left: Dyn runs. Middle: Bz50 runs. Right: Bz200 run. Bottom panels: Correlation coefficient between temperature in the “ambi” and “clean” runs as a function of time. Vertical lines mark times where the correlation drops to 95% (dotted lines) and 90% (dashed lines), where appropriate. We note the difference in the axes range between the middle panels compared to the left and right panels.