They base their numbers on reported dates. So it seems like as deaths “spike” the states have a hard time keeping up with the reporting. So in July Worldometers was undercounting deaths, but they were going up so it was still all fear porn all the time.
Then when they started going down, Worldometers was getting all deaths they missed from back in July so the decline in August/September didn’t look as sharp as it actually was. So they can keep the fear going better than if the true curve was really being represented.
Even as deaths have risen in October, they’ve managed to be higher than the CDC which goes off date of death rather than reported date. Remains to be seen if the CDC is just behind or if Worldometers is continuing to “harvest” old deaths from the spring and summer.
I believe a similar issue is going on with “cases.”