That's exactly what I don't understand. Logically, wouldn't deaths typically lag diagnosis by some amount? If so, we would expect the death rate to be higher than current deaths/cases because we need to use the confirmed cases from some number of days ago. That number of days being the average time between someone being diagnosed and dying.
Another stat is that there are roughly 700 deaths and 2400 recoveries, which would indicate a mortality rate of 23%. Of course, this suffers from the same delay issue as the 2% calculation (T_death < T_recovery). But the possible range is 2% < M_rate < 23%