A bias-aware analysis scheme designed to correct bias in either the background or the observations will reduce mean analysis increments by construction, but not necessarily for the correct reason. It is necessary to test whether the analysis has improved by verifying that the bias attribution is accurate. Figure 7 illustrates how a successful bias correction of the background during assimilation should lead to better analysis and reduced forecast errors. However, reducing the bias in the initial conditions may only improve the forecast in practice if the model itself is changed.