People want to use CFR as a measure of how worried they should be, or as some kind of comparison between countries, and it's totally inappropriate for either purpose.
The statistics always seem to go up no matter what kind of interpretation.
When the health system is severely overburdened, that subset because a smaller and smaller proportion, as scarce resources are used for testing of people presenting at hospitals rather than population-based testing.
People who are asymptomatic or have minor symptoms don't present at a hospital, so they don't get tested, so they never get recorded as a case (which, by definition, requires a test and a diagnosis).
The real population fatality rates are guaranteed to be lower than the CFR unless you believe that every person who is infected is being tested and diagnosed, and it's my opinion that they'll be much lower.