[0] http://www.reddit.com/wiki/faq#wiki_how_is_a_submission.27s_...
The FAQ you linked to gives this example: a submission with 5 upvotes and 3 downvotes may be tweaked to say 23 upvotes and 21 downvotes.
So the net number of upvotes remains the same. But the ratio of approval, a statistic that is prominently listed for each submission, changes:
5 / (5+3) = 62.5%
23 / (21+23) = 52.5%
I thought I had also read that, in earlier implementations of the fuzz, upvotes were either removed, or downvotes added, so that doing an all-time view of popular posts would not completely consist of new posts (because the reddit user base is much bigger than it was several years ago, so a relatively mediocre post could garner more upvotes than a very good older post).This is the entire point.
That said, I wish they'd just ban spammers and give everyone accurate metrics.
The best they can do is make spammers be unclear on if they're having any effect or not, by obscuring votes, or by secretly banning them without them realizing.
It's a tough problem to get right, especially without making the experience worse for everybody else. I think they've found a happy medium.
... for crying out loud.
>That said, I wish they'd just ban spammers and give everyone accurate metrics.
As long as spammers are banned, there's no reason to screw up a site feature for everyone else.
That's already almost half the net upvotes of what Obama got: http://www.reddit.com/r/IAmA/comments/z1c9z/i_am_barack_obam...
But Obama's IAMA got 239,221 upvotes total. Which means he got 225K downvotes...which I guess is feasible, but for all we know, that was just an arbitrary result Reddit's strange algorithm of adding downvotes to obscure the upvote count.
I guess it doesn't matter in the end, it just seems that upvotes should be a real reflection of how liked (or polarizing) a thread/discussion was.