Moderation is not easy, but it's not
that hard either. Actually, it has already been figured out. It's partly what the legal system is about. We also know a few online forums that work just great (or, you could say well enough). This includes HN, but it's not the only one. I happen to moderate a few rather large FB groups (about 50k people in 3 groups). Now while they are not too active, it's pretty obvious what works. And yes, one of them is transparency. Just as with the legal system.
And this is one of the things that are broken on the large social media platforms. I mostly use FB, so that's what I know about. But what they do is so ridiculous, that I started to collect their decisions. I've been banned for a week for calling a politician, a public figure (with 50k followers) "braindead" in response to a really stupid comment he made publicly under news article. (That's all I said. It's a widely used expression in Hungarian for saying that someone is an idiot. It's the exact same term as the medical one for someone being in the state of brain death. Not exactly a nice thing to say someone but not over the top either.)
He used his public profile (not his personal one, but the FB page). I got blocked 20 minutes after making the comment for harassment . Their policy explicitly says that public figures should accept more than ordinary people. And this is what the law says as well. At the same time I keep reporting people making homophobic, racist comments or simply just being flat out rude, using obscene remarks (addressing me or others) and they almost always get rejected after 3 days.
But since transparency is non-existent, no one really knows what's acceptable and what's not, what the consequences are (it seems that the penalties get longer and longer, doesn't matter how serious the violation is, but there might be an expiration, so maybe only the last year counts).
At the same time what I found to be working (which is the same I see here) is that instead of deleting, which is inefficient for a lot of reasons, you should issue warnings, explaining what is not OK with the comment. This seems inefficient at first, but it works because it adds transparency and helps everyone understand and internalize the rules. It also helps people to calm down because they'll see that if others misbehave they won't get away with it either.
But yes, algorithms don't work (yet), especially not in non-English languages. Underpaid, invisible, who-knows-whos might be cheap, but they don't cut it either.