I understand that the details of the vote anomaly detection script are secret for a reason. But I also think that at least the coarse principles of what amounts to automated policing of what is legitimate voting behavior should be up for discussion. And many users might not even be aware that such as script exists.
Personally, I essentially only down-vote questions and answers that I consider to be in bad faith. If I see a mathematical mistake I consider not to be the outcome of stubborn crankery, I leave a comment and do not downvote. I do downvote when I have the impression that someone violates the community norms while knowing better. If someone posts ten questions in a row that seem to come from some problem-set without showing any effort, even though the poster has been asked to modify their behavior, I feel justified in downvoting the whole series of questions. If some Cantor-got-it-all-wrong crank starts giving "answers" based on this mistaken view, I feel free to downvote all these answers. This can be very fast, and it might be perfectly possible that this looks like spiteful-downvoting to an observer who is not able to form sound judgements based on mathematical knowledge and knowledge of human thinking: I'm talking about a machine here. So I have the following questions, which I think can be answered without endangering the integrity of the system:
- Are the actions of the vote anomaly detection script reviewed by moderators?
- Is my voting behavior likely to trigger the vote anomaly detection script?
- Can somebody tell me whether the vote anomaly detection script has ever reversed votes of mine?
The only, and rather controversial, discussion of the vote anomaly detection script I have found is in the comments of this post of Jeff Atwood.
Addendum:
I've found some information by looking at stackexchange proper. There is this blog entry by Jeff Atwood when the system got introduced. There is a lot of discussion following this post. A useful summary for thos who do not want to wade through the drama can be found here. Apparently, developers and no-one else have access to individual voting data.