All calculations are stable broken-symmetry solutions close to the complete basis set limit, and have been performed using quadruple‑ζ quality basis sets using Q-Chem 5.1 \cite{shao_advances_2015}. Results for the three statistical criteria, AIC, VCC, and CVC, are reported in Table \ref{182001}, as well as the ranking of each functional according to each specific criterion (in parenthesis). The average ranking of each functional across the three criteria is also reported in the last column of Table \ref{182001}, and is used as the final indicator for performance of a functional. It is clear that the rankings obtained using the statistical criteria align well with the Jacob’s ladder picture of functional approximations. According to all three criteria, for example, the three best functionals are double-hybrid “fifth rung” approximations. “Fourth rung” hybrid meta-GGA/NGA are the second-best class, followed by “fourth rung” hybrid GGA/NGA and “third rung” local meta-GGA, with similar average performance. “First and second rung” Local GGA/NGA are on average at the bottom of the rankings. Interestingly enough, modern non-fitted functionals such as PBE and SCAN-D3(BJ) sits in the middle of the ranking, together with most of the parametrized Minnesota functionals. Even more interesting than the general trends are some of the outliers. For example, the B3LYP‑D3(BJ) ranks near the top according to all three criteria, while its parent functional B3LYP is consistently ranked at the bottom, more than 23 positions below B3LYP-D3(BJ), confirming the trends observed in the literature. However, PBE-D3(BJ) is slightly more transferable (despite a slightly higher wMUE) than B3LYP-D3(BJ), confirming recent finding of transferability issues in the popular B3LYP-D3(BJ) functional.