I get my data from the
Worldometers page for the USA. (there's even the option of seeing the raw data overlaid with the 7 day average)
I think the Johns Hopkins graph of the 7-day average is wonky - it looks like the calculation may include future non-reported numbers for the very latest days - it looks like all their 7-day averages drop off at the end.
It's better to look at the raw data - for instance, if last Wednesday's numbers were less than this Wednesday's numbers, then the 7-day average would necessarily have gone up in the past day.
For instance, here are the past 15 days:
Date Day New Cases
2020-12-09...
Wednesday...2254412020-12-08...Tuesday.......211011
2020-12-07...Monday........200572
2020-12-06...Sunday........183358
2020-12-05...Saturday......212990
2020-12-04...Friday..........238065
2020-12-03...Thursday......220771
2020-12-02...
Wednesday...2063792020-12-01...Tuesday.......184169
2020-11-30...Monday........167033
2020-11-29...Sunday........144644
2020-11-28...Saturday.......146631
2020-11-27...Friday..........166059
2020-11-26...Thursday......161411
2020-11-25...
Wednesday...183596