Mean redshift calibrations with the network Net_P. Upper panels: residuals of mean redshifts (different from mean residuals) in each z_photo tomographic bin for the baseline method and step 4 of our methods using the network Net_P with the SDSS data, the CFHTLS-DEEP data, and the CFHTLS-WIDE data. A magnitude cut r < 24 is applied to the CFHTLS-DEEP data to create a restricted sample. The shaded areas indicate the requirement for the accuracy of photometric redshift estimation (i.e., |Δ_z| < 0.002). Steps 1-3 are applied in advance of the calibration by step 4. Middle panels: same as the upper panels, but residuals of mean redshifts are estimated in r-band magnitude bins. Lower panels: 1-Wasserstein distance between the z_photo and z_spec sample distributions (Eq. (13)) as a function of r-band magnitude.