Independent check of neural network ANNz photo-zs, z_Reis, against EAZY photo-zs (left) and BPZ photo-zs (right) for a galaxy sample with no other redshift information available (roughly 30% of the UKWIDE sample across all redshifts). Catastrophic problems in z_Reis resulting from inadequate training sets are not evident. Hints of potential photo-z biases at the 0.1 level follow different patterns in the two panels and are difficult to confirm because, by definition, no spec-zs are available for this selection.