5

While I turned down a similar request on Super User, our tooling and circumstances are very different - and in discussions with a Server Fault moderator, the prospect of having additional automatic flags (up to 4) with a human in a loop for the final flag came up.

It might be a good idea,for maybe a month or two (or as the folks directly involved see fit), to help mitigate spam waves with additional assistance this way for posts with a high certainty of being spam - since SF dosen't currently have the same core of regulars, nor the tooling we've set up on SU over the past two years for now.

1
  • 1
    Marking this deferred, and will review again once the upcoming anti-spam tooling has been live for a while to determine if a change like this will be necessary. Commented Oct 2, 2025 at 21:19

2 Answers 2

Reset to default
4

At the moment that's the maximum for automatic flags, as to do more would remove posts without a human viewing it first. "Autonuking" posts has been discussed before, and as far as I know the decision has always been not to do that.

My view of the situation is that it's not necessary to autonuke posts to get rid of the spam. I think other measures could help, like increasing the number of flags available to each user, as discussed recently here and here. My guess is that StackExchange could probably do more on their end, but I wouldn't know what or what the tradeoffs would be.

1

Currently 3 autoflags is the amount used. It takes 4 flags to delete a post, so there is currently 1 human in the loop, but not more. There is basically no point in casting more autoflags than this, as long as you are unwilling to remove that 1 human in the loop.

However, I do think that we should go to 4 autoflags and autonuke posts that are certainly spam.

It is easily possible to make SmokeDetector conditions which catch a lot of spam and are perfectly accurate. For example, in all of the time SmokeDetector has existed (since 2016), there has not been a single false positive with more than 5 "reason weight" and more than 4 detection reasons. But such posts encompass 49.78% of all spam that has been posted to this site.

We can do a bit by adding another condition, and autonuking any post with more than 273 "reason weight" (again, this condition has perfect accuracy). Between these two conditions, we've now encompassed 50.62% of the spam on this site.

But, we can also encompass more spam in our perfect-accuracy autonuking by using keywords. For example, 34.03% of all spam posted to this site has contained "helpline" in the title, while no non-spam post ever has. Of these posts, about a quarter (there isn't an easy query for this, and I don't want to go through all 4k+ posts manually, so this is based off extrapolating from a random sample of a few hundred of the matching posts) wouldn't be caught by our 2 conditions from before, so about 8.51% of all spam. Adding that to the conditions above, we get 59.13% of spam that could be easily autonuked.

Given that we can clearly get rid of the majority of the spam posted to this site automatically, without the spam staying on the site until someone flags it, and without people needing to interact with the spam, why wouldn't we do this?

Frankly, I'd imagine we could probably have better conditions than these, that would catch even more spam with perfect accuracy (after all, I came up with these in 5 minutes).

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.