On: New Anti-Toxicity Features on Bluesky

[perihelions]

I think you can estimate the threshold by just counting the ratio of times you flag something and it immediately dies. You would wait a while to see which items die or do not die, and discard the subset which never die, and consider only the subset of items that transitioned to [flagged][dead].

If the [flagged] logic is simply "flags >= N", then of the subset of samples that are flag-killed with your involvement, you will be final flag in 1 of N of those.

The null hypothesis is "this experiment converges to a reciprocal integer".

[runarberg → perihelions]

I feel like this requires an assumption that any post you flag is equally likely to be flagged by other users, and at a similar time as you, in other words that your flagging behavior is very similar to others that use this feature.

I guess this might give you a pretty nice upper estimate though. Unless N is very high / very few users use the feature, and you are frequently much later to the scene as the average user of the feature. Then N may be underestimated.

[krapp → runarberg]

Hacker News is biased so that the threshold for flagging is very low, in order to maximize the effective separation of signal from noise as much as possible (for as broad a definition of "noise" as possible.) If I had to guess it's probably 3, or it's tied to user karma (flags from higher karma accounts count more.)

And certain topics are likely to be flagged by multiple people, this can be assumed based on the topics that people complain aren't "HN worthy." Anything that can be considered politics, for instance, is probably going to get flagged because of the number of people who don't believe political stories of any kind have a place here.

[PaulHoule → krapp]

Topics about US electoral politics get flagged consistently, as does anything having to do with sex, sexuality and gender.

Amazingly I see articles from Trotskyite populations often don't get flagged because, oddly, Marxism doesn't seem to be political anymore.

[runarberg → PaulHoule]

This post was my first in this experiment. I flagged it when it was 10 min old and was already downvoted at least once.

PS. @PaulHoule. I flagged it for valid reasons, not just to conduct this experiments, and not because it is political.

[PaulHoule → runarberg]

No prob. It is not like I can’t afford to lose a few points.

[hansvm → perihelions]

That's probably a decent approximation, but it assumes the distribution of latencies for flagging is the same for you and the aggregate "everyone else."

[PaulHoule → perihelions]

I think so. It depends however on how fast I think I am compared to the other flaggers. If I was really, really fast I could always be the first flagger.