I figure there's some kind of "true" fatality rate that just answers the question of "If you catch the virus, you have this (age-adjusted) chance of dying". It seems like we'll asymptotically reach that "true" rate once testing is extremely common and liberal (to have confidence you are identifying all cases). Some countries are reaching that testing rate. Then your upper and lower bounds for that death rate are "deaths / deaths + recoveries" and "deaths / confirmed cases", respectively, and they should converge as "active cases" decrease over time (as they either recover or die). You can use that "band" (between the upper and lower bounds) to judge other countries and determine whether they are testing enough - if they are above that "band", they are clearly not testing enough.