TB12 series top, TB13 series bottom - test is opening a new window (target about:blank) (which = new tab thanks to our prefs) and measures as many times as possible, as fast as possible, until we get an error - it then analyses those measurements and records any changes
13.0a1 leaks (as per pic) . note 13.0a2 jumped newwin size to 1400x900, so tested version in OP is definitely 13.0a1. note to self: state version tested in posts, you have gained -5 intelligence
13.0a2 seems fine as well. tested with and without bookmarks toolbar, and with a wrong newwin size (e.g. drag slightly taller/wider, so letterboxing was used) - but these have nothing to do with the issue (except to change the outer value if leaked)
any non-default preferences
nothing of note: I usually have bookmarks toolbar to always show in my test suite but that's about it
So seems it fixed itself as the result on something in 13.0a2 (would love to knwo the reason WHY it failed). We really need more robust detection of failures - I don't always click manual tests on TZP, just lucked on it TBH