Estimated photo-zs vs. true redshifts for the source galaxies in our simulated sample. The photo-zs were obtained by running the BPZ algorithm on the best-fit colours shown in Fig. 4. The black line shows where the photo-z equals the true redshift. The grey shaded region indicates where δz/(1 + z_true) < 0.15 (i.e., where the estimated redshift is not an outlier). The outlier fraction (fraction of the cases with δz/(1 + z_true) > 0.15) is 8.3% and the 1σ scatter of redshift errors (δz/1 + z_true) is 0.032. As in Fig. 9, the redshift with the highest probability in the redshift PDF was used as the estimated photo-z. The error bars in this figure correspond to the 68th percentile of the redshift PDFs (grey shaded region in Fig. 9.)