Does it take into account gross upvotes, or net?

E.g. could a controversial thread (plenty of upvotes, plenty of downvotes) get pushed off the front page because it has a low upvote:comment ratio?

I get that YC probably don't want to deal with controversy, but, well this is a community, of sorts...

I sometimes wish some stuff would stay up, but, in the aggregate, I'm pretty happy with the result.

Whenever I see some article posted, that is likely to generate a storm of bickering (many platforms elicit this; not just politics), I'm usually happy to see it take a nosedive.

I had something that I submitted, fairly recently, get plonked. It was actually fairly sensible and relevant, but one of the commenters posted something that was "sort of" a "dox," and I understood why it disappeared.

> E.g. could a controversial thread (plenty of upvotes, plenty of downvotes) get pushed off the front page because it has a low upvote:comment ratio?

You can't downvote submissions, just flag them. As for comments, I'm not sure how the sorting works there (but it seems like net upvotes, with comments younger than an hour being pushed to the top).

> gross upvotes, or net

these two are the same since you can't downvote submissions.

I couldn't think of a better criteria of success when automating something.

Only that more than half of the time a peaceful discussion gets buried by that automated system and then dang comes several hours later and says "oooops this got caught by the anti-flamewar system and now the discussion is dead!! oops!! our code monkeys are working hard to fix this!"

Where's the data to support that opinion

What opinion? It's a statement of fact, not an opinion. The flameware detector is a well-documented and old HN feature:

https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...

All the data is available at https://news.ycombinator.com/.

As a suggested experiment, randomly select 25 stories with comments>votes, and 25 stories with votes>comments. Go through the comments on those 50 stories and score them for flamewarishness. It's probably a half day's work. 50 stories * 3 minutes each for scoring, plus some time for the random sampling and stats.

You could present the data as a plot of your subjective flamewar score vs. log(votes/comments)

Where’s the data to support the opinion that opinions should be supported by data