Photometric redshift quality for the faintest galaxies in the S82-MGC. Results using the template-based EAZY (left panels) and BPZ (right panels) photo-z codes are compared to VVDS and DEEP2 spec-zs as indicated. VVDS is a magnitude limited survey with i ≲ 22.5, while DEEP2 targets were color-selected to have z ≳ 0.7 and typically have magnitudes of i ≳ 22.5 which therefore reach or exceed the S82-MGC detection limits. Catastrophic outliers display significantly different behavior between the two codes. Given sparse training sets at faint magnitudes, the ODDS parameter output from EAZY and BPZ provides an advantage over neural-network photo-zs for selecting better performing photo-z samples. For example, the lower panels show improved results for ODDS > 0.7. The percentage of the original sample that satisfies the ODDS cut is labeled in each panel as are metrics characterizing the resulting photo-z quality.