Particularly with data set #12 and #13. See https://travis-ci.org/wikimedia/mediawiki-core/jobs/24510844#L352 on travis
The method gets most of the date correct but gets the day itself slightly wrong.
To me this looks like something that changes slightly from system to system.
The tests (or these 2 in particular) should probably be changed to ignore the changing day and just test the rest of the timestamp.