Perhaps it's better to reference scores to standard deviation rather than the full possible range -- I guess it depends on the particular case.

Yep, that highly depends on the shape of the distribution.

One Gleb Meaningfulness Metric point, likewise, is just a certain fraction of how much variation is possible

Well, we (or at least I) haven't seen Gleb's Meaninfulness Metric, so I have no idea if it's defined via population standard deviation like IQ. It may or it may be. I brought up IQ as an example of a unit which does not directly correspond to, say, thinking speed or working memory capacity -- it's entangled with the test itself, but it does make the numbers interpretable.