And how you know everyone infected will be cured if I go by your logic?
For real time events the calculations are done as per available facts not assumptions. As in your case if everyone infected starts dying we are grossly underestimating fatality rate, so available data is used for approx estimates.
Actually you don't need to know how many will get cured before hand. Not at all. You
cann't know it.
Whenever you calculate currently (what you call "real time"), you will always overestimate the fatality rate because of three` simple reasons :-
1. Deaths due to Covid in --industrialized countries atleast-- will HIGHLY LIKELY to be reported. Why? Simply because death is permanent unlike sickness which goes away. Those who are really really sick (and hence they end up dead) will get tested surely --atleast in countries like Italy, Japan, Korea-- and their deaths will be correctly attributed to coronavirus. Net result, your death data attributed to coronavirus will be pretty bang on.
2. Those who are mildly/moderately sick are being less likely to be tested, EVEN in countries like Japan, Korea, Italy, Canada. Simply because most of these places even now is rationing medical resources to vulnerables and badly sick people as these resources --testing kits, medical personnels time etc. Those who are/were mildly/moderately sick will get symptomatic medicines and are very likely to get well before they are even tested, if tested at all. Meaning, your number of people who are infected are LIKELY to be underestimated.
3. Vast majority of people who got coronavirus develop mild and moderate symptoms(WHO confirmed this). So, again as a corrolary of (2), you are vastly underestimating the number of infected folks.
Hence, your fatality rate, dead(bang on)/infected(underestimated) will be an OVERESTIMATION of reality.
Now once this disease is over and people will do deeper analysis of covid and find more accurate estimates of infection numbers, we will actually know the real fatality rate. Currently the fatality rates estimations are very much likely to overestimate the actual rates.