That would be my assumption as well. It appears in the "public data download" they are backdating the deaths to the date the death actually occurred. In their tweets, they identify the number of deaths reported for the day, rather than the number that actually occurred.