All good input so far.
Furthermore, to piggyback on Val’s last bullet point, I’d add that such tests should show a high level of test-retest reliability; be sufficiently generalizable outside the calibration sample so as to produce reproducible results; be tested and recalibrated in large, long-lived panel samples that are representative of the general population and; where possible, be examined in twin studies against a vector of descriptive variables to control somewhat for confounding environmental factors.
