Yes the basic principle has changed a bit, albeit longer ago, start 2017.

In preparation for Wikistats-2 one Wikistats-1 metric has been updated: edit counts now also comprises edits on redirects. Redirects were ignored earlier primarily to not make article counts explode (wp:de has 2.1 M official articles and 1.4 M redirects). For articles counts this filter does still apply. But it totally makes sense to honor the effort of creating redirects and make this count towards a users activity.

But this change happened in December 2016. gitub:WikiCountsInput.pm, line 905:

if ($content_page) # Dec 2016 redirects are no longer exluded from edit(or) ocunts, they are still for article counts

I don't have reports or raw data for Aug vs Sep, but Internet Archive stores periodic copies of Wikistats reports. The following comparison draws on data from WaybackMachine, comparing runs from Oct 17, Jun 17, Feb 17, Feb 16, Aug 15:

Please note that new editors (column B on wiki specific stats page, first table) is a trivial derivative from column A (total wikipedians), simply the month over month increase. So my analysis focuses on column A. As you can see the delta A (difference for some month over consecutive Wikistats runs is very close to 100% (tiny increase is expected as more users reach the threshold of 10 edits, which is reported at the month of their first entry).

Feb 16 to Feb 17 shows an increase of almost 2% for every historic month. That must be the Dec 2016 patch.

However the difference in Dr. Bernard Gross's chart, seems more in the order of 10%, at least for peak value Jan 2007. That I can't explain, and it doesn't show in the reports I analyzed, which are pretty consistent except for the expected increase after Dec 2016 path.

Sorry I don't have recent backups for every month of StatisticsMonthly.csv. The dumps process stalled after migration to new server in July (I'll work on that).