Greg KH suggests the following in pull request 1, #1 :
"Tests for major feature additions as without some kind of test being present, it's hard to verify if something new even works. This is now the "unofficial" rule for the kernel, as we have been burned by this in the past (feature additions that never even worked)."
I like the idea of adding this to the criteria set, but I think projects need to make it "official" in two senses: (1) their docs on making changes need to require it, and (2) there needs to be some evidence that they're doing it. If it's a super-secret rule, or one not actually used, there's not value. I don't want to add criteria to the "basic" list that few projects meet, though. If this the "unofficial" rule for the kernel, would they be willing to do that?