# Does the vote abuse algorithm roll back valid votes?

My understanding is that if you vote up (or down) too many posts by a single person, those votes will be reversed as abuse. This is to prevent attacks on another user by downvoting every single post they've made and to prevent sock-puppetry when someone upvotes every post a friend has made.

Fine, but on SF (where I vote a lot) there aren't a huge number of regulars and many of them give good answers to lots of questions. Just the other day, the guy with the most SF rep answered at least 10 questions with his usual good answers. Knowing that votes will be rolled back if I upvote too many of a single person's answers in one day, I try to avoid it, but I'm pretty sure that over the past few days it's happened a couple times.

So I'm wondering:

• Is there a limit up to which it's safe to vote for a single person? e.g. don't cast more than 5 votes/day (or 10/week or whatever) or they might be rolled back.

• Or, could the algorithm be changed so that it doesn't roll back legitimate votes on a single person's answers?

• I doubt you'll ever get the answer to your first question since revealing that information would allow the fraud algorithm to be exploited to avoid repercussions for fraudulent or malicious voting. – Adam Lear Jan 28 '12 at 20:25
• What characteristics could be used to distinguish between legitimate votes and fraudulent ones? The system can't know. All it can see is number and frequency it can't see intent. – ChrisF Jan 28 '12 at 20:51
• @AnnaLear: With a binary splitting algorithm, I could start by 16 votes on a day, and find iteratively: 16-8-12-10-11 all limits below 16 in 5 tests. Higher limits would need more steps, 1 more step per power of 2. – user unknown Jan 29 '12 at 1:15
• @userunknown It is probable that if you were caught by the fraudulent voting algorithm more than once, something would happen. – kiamlaluno Jan 29 '12 at 12:42
• @kiamlaluno: Of course, for such actions, I would always use one of my sockpuppets. – user unknown Jan 29 '12 at 13:35

Or, could the algorithm be changed so that it doesn't roll back legitimate votes on a single person's answers?

No, it can't be.

Stop and think about this for a minute: what's the difference between a "fraud" vote and a "legitimate" vote?

A: The intentions of the voter.

There isn't an algorithm - or indeed, a person - qualified to judge the intentions of a person voting. Sure, we could tweak the workings of the fraud detection algorithm to help catch likely fraud while still avoiding penalizing legit voters (and we have, quite a few times, which is why you're never going to get a valid answer to your first question), but at the end of the day it's still a poor attempt at solving an impossible problem. For every person who gets unfairly penalized, there's someone else who feels that penalties should be quicker and more strict.

The best advice I can give you is simply: be equally free with your votes. If you're using x votes on user A, you'll ideally be spreading x*3 votes across users B, C, and D - maybe not up-votes of course, but some roughly equivalent number of votes. If you're limiting the bulk of your voting (again, up or down) to one other person, that looks bad - and even if you have totally legitimate reasons for doing so, they may well be misinterpreted.

• Sounds conclusive. Thanks. :) – Octavian Damiean Jan 28 '12 at 23:04
• Not so easy when A is particularly prolific (ask Jon Skeet). – Gilles Jan 28 '12 at 23:57

Since your first answer is unlikely to get answered and the second one is pretty much impossible to implement I would propose this.

Up-vote some of those answers, mark the other ones that would deserve up-votes as favorites and come back the next day and vote those up as well.

• Excellent new loophole to check for in the fraud algorithm; thanks! (^_^) – Awesome Poodles Jan 28 '12 at 21:05
• Well, we are not really helping the community with making it impossible to honor active users that post awesome content. – Octavian Damiean Jan 28 '12 at 21:06
• It's a nice idea, but... I gotta tell you, there's absolutely no guarantee that this technique will keep you safe either. The only "safe" way to vote is: don't vote for only one other user. – Shog9 Jan 28 '12 at 22:57
• @Shog9 That's not what I said the problem was. I said there are times when it's totally appropriate to cast 10 votes in one day for one person, and I don't think those votes should be rolled back. It's like a penalty applied to people who answer well and frequently. – Ward Jan 29 '12 at 3:30
• @Ward: And I'm saying, there's no algorithm that can judge your intentions. We've decided (some three years ago now) that it looks dodgy when one person spends a lot of votes on another - the exact parameters of this test are adjusted to try to compensate for natural and expected behaviors, but the core assumption - that honest users generally spread their votes around a bit - is hard to challenge (even allowing that there are valid exceptions). – Shog9 Jan 29 '12 at 20:09
• @Shog9 Then let's get to a case that shouldn't reveal anything about the algorithm: have I had votes reversed in January for voting "too much" for one person? Or in the last quarter of 2011? – Ward Jan 30 '12 at 5:21
• @Ward: So far as I can tell, no. – Shog9 Jan 30 '12 at 15:55