> So it seems like they should ignore pubdate as it could be gamed, and just measure the real "last updated" date based on content changes.

It's hard to imagine a mechanistic measure of the 'real' last-updated date that couldn't also be gamed. (If I'm being really cynical, it's hard to imagine anything that couldn't be gamed; so I guess I mean 'easily gamed', for some value of 'easily'.)