I have to say, I've always admired the Stack exchange system. Yes, it's a Karma-like system, and it's obviously not perfect, but it means that accounts always start with very little abilities, most notably that they're not able to downvote yet. And when those accounts do get the ability to downvote (which doesn't come all too quickly), it costs a certain amount of their "reputation", which makes them think twice about downvoting.

Gog disabled down votes on its forum and now there's a bot up voting every reply in derailed threads. Mass up voting can also be a problem in creative hands.

I suppose that would address only a part of the issue and there are other, less intrusive ways to mitigate the effects of malicious early down voting. For instance, early down votes could be weighed less.

PieFed, at the discretion of community mods, offers restriction of voting to only subscribed community members. This limits drive-by downvoting from All, where people would not have read the community rules (which in PieFed are repeated in their entirety at the bottom of every post from that community).

It also offers restriction of voting to only "trusted" instances, thereby introducing a third category between the binary federation vs. defederation.

I have also seen communities on PieFed that disable downvoting entirely, even to subscribed members, even on the same instance.

Community mods can enable or disable these settings at will iirc.

For positive voting you could look at how quickly accounts upvote after a post has been made, combined with how new they are, and whether they have comments or not (maybe also if those comments seem AI-generated).