Generally there is an overall agreement in most aspects of photometric redshift methodologies, and even technicalities. However there is a need for standardized quality measures and testing procedures. It is important to analyze to performance of each model spectrum as a function of the redshift. This is best done by plotting the difference of the observed magnitudes and template-based ones. These figure can pinpoint problems with the spectra and even zero-points. Intrinsically these are the quantities used inside the template-optimization procedures, e.g., in Budavári et al. (2000) and Feldmann et al. (2006).
In SED fitting, interpolation between templates is often used, which can be done linearly or logarithmically. The latter has the advantage of being independent of the normalization of the spectra. Yet, most codes appear to use linear interpolation without a careful normalization. This might explain some of the discrepancies among similar codes found by the Photo-Z Accuracy Testing (PHAT) project. 5
The determination of the quality of the estimates is also a crucial topic. There is need for different measures that can describe the scatter of the points without being dominated by outliers and that can estimate the fraction of catastrophic failures. It is also recommended to characterize the accuracy of the estimates by a robust M-scale instead of the RMS; a measure that is simple to calculate, yet, not sensitive to outliers. Another aspect of this is the study of selection criteria that is often neglected. Certain projects are not concerned with incomplete samples as long as the precision of the ones provided is good (e.g., weak lensing, Mandelbaum et al. 2005), while others, such as galaxy clustering, might require an unbiased selection. Therefore, it is perceived that studies using methods with any quality flags or quantities should provide details of their selection effects.
A common theme for future goals in most photometric redshift works appears to be more detailed probabilistic analyses, with the need for probability density functions. Priors used in most bayesian analyses seem to be generally accepted in the photo-z community. With such consensus amongst photometric redshifts obtained, the focus of work now is shifting from the estimation of “just” the redshifts to simultaneously constraining physical parameters and the redshift in a consistent way.