If the flamewar detector takes content in to account, could it consider the overall tone of the discussion and the content of each new comment and selectively kill new comments that seem flame-like?

I wrote a moderation bot to use on a political subreddit that used a text classifier to determine whether comments should be deleted. A goal was to delete flame comments rapidly. It worked, but reddit's reply notifications limited its effectiveness.

Of note: the phrase "you are a" ranked higher for the flame category than any particular insult.

Ahh, that makes sense. Can there be a sliding scale of delay for the ability to comment, or depth of thread allowed (before commenting is disabled for that thread), based on the output of the flamewar detector?

This may also have the side effect of allowing for some opinions to be shared, without being mired in flame-y back and forth. (And, importantly, still allows the story to stay more visible)