Token / translation model has a fuzzy value which is mostly unused.
Since we will soon have a "canonical" set of full texts for all sites, making decisions on retiring old tokens should be easier. Currently unused tokens should enter a "dormant" state where they are still fully accessible from the translation system, but aren't counted towards coverage or published in locale packages.