Skip to main content


inspired by @darius's talk at #fossy24 about the study he and @kissane did about fediverse moderation, i figured i would outline how treehouse moderation works.

so if that does not interest you, feel free to mute this thread.

in reply to Ariadne Conill 🐰:therian:

first off: screening new signups has helped us a lot.

when we first launched treehouse mastodon, because we have high profile community members, we attracted trolls basically immediately whose goals were to disrupt operations of the instance.

the most notable incident involved a person who signed up for an account and then immediately uploaded multiple pieces of media which were CSAM.

this actually has wound up going very badly for that person, who has had to cancel his podcast several times because his door has gotten kicked in by law enforcement due to this, and other maladaptive behavior.

in reply to Ariadne Conill 🐰:therian:

when folks complain about getting crapflooded -- their complaints are real.

as has recently been highlighted -- in the case of BIPOC, these crapflooding trolls tend to flood users with racial slurs or similar.

in the case of non-BIPOC, these crapflooding trolls tend to use other angles of cheap low-effort insults in their crapfloods.

but crapfloods cause users to have a bad experience, regardless of what the flavor is.

we have built custom tooling to deal with various styles of crapflooding attack, by simply hacking automated mitigations into the mastodon codebase. there are a few feature flags on the treehouse mastodon fork to turn on and off those mitigations.

mitigating crapfloods with automation is essential to ensure a moderation team can scale. otherwise the moderation team will spend a lot of time cleaning up crapfloods.

defederating typical sources of crapfloods is also recommended, but we have dealt with trolls who farm accounts on normal instances that wouldn't typically be defederated.

so you really need automation here.

in reply to Ariadne Conill 🐰:therian:

in the case of treehouse specifically, the majority of the treehouse staff have prior experience managing and moderating IRC networks. this, of course, gives us a good understanding of tactics that can be used to mitigate crapfloods.

basically everything that applies to mitigating IRC abuse applies to mitigating fediverse abuse. bucketing and other basic trend analysis techniques provide sufficient signal quality to do automated moderation against.

in reply to Ariadne Conill 🐰:therian:

the next area is obvious: limit vs defederation

in general, we just defederate instances that become sources of annoyance to us.

in some cases we will limit an instance, such as mastodon.social, in response to those instances being a source of spam, but this is pretty rare, because almost all of the instances we do instance-level blocking against are run by trolls who think flooding people with nazi shit is somehow new and novel.

of course, when you've been moderating IRC servers for 20 years, none of this shit is new or novel. it's just boring. and so we defederate it.

in reply to Ariadne Conill 🐰:therian:

when it comes to *being* defederated, it isn't the end of the world.

some people have a different style toward how they want to run their space. although they may see their defederation of *you* as a value judgement, it probably isn't in reality: they probably just find the friction annoying.

what we've learned is when staff are seen as brooding about being defederated from an instance, it results in further defederations.

learning to not feed into this is essential. you must, as an instance admin, maintain command of a situation, and that involves checking your own shit.

there are a number of instances that have defederated treehouse. it doesn't matter.

focus on *your* community. users sign up for *your* community because they want to be part of *your* community.

if they wanted to be part of the *other* side of the netsplit, they can sign up for an account on that side of the netsplit.

yes, this doesn't feel good, but ultimately not feeding negativity is key to avoiding getting burned out over "fediverse politics."

in reply to Ariadne Conill 🐰:therian:

the larger fediverse is best seen as "bonus content."

it's great that you have access to it, but your community should be able to stand on its own merit.

if it can't, then why are you doing this?

in reply to Ariadne Conill 🐰:therian:

the last thing i will say about politics is -- stay in your own lane.

nothing of value will ever be gained by inserting yourself into somebody else's dispute.

the few times that we have ever done this, we won a stupid prize -- defederation from one or more instances.

it's just not worth it, and it's not your business anyway.

in reply to Ariadne Conill 🐰:therian:

now, there is the issue of managing moderation burnout.

the solution here is: recruit more moderators.

in the case of treehouse, we have taken inspiration from the days of running EFnet IRC channels, and the War of 1812: if somebody looks like they know what they're doing, they're pressganged into service.

however, you can't just pressgang someone into 24x7 service as a moderator.

so building a system where carrying the pager is load balanced is essential. community members should not have to be expected to wear the mod hat all of the time.

there are a few options here. you can formally set schedules, or you can just set the community norm that participation in moderation is best-effort.

it would be cool if mastodon integrated with a tool like pagerduty directly, for example.

internally, we have some hackish tools that enable and disable moderation alerts to allow folks to take breaks as needed.

coverage -- to allow rapid response to incidents -- is also essential. in treehouse, we have built up a robust enough moderation team that there will basically always be somebody around, no matter what time it is.

this is because we followed the IRC model of building a mod team -- the more, the merrier.

but you can't just make every user a mod and hope that works. finding the right balance is very much dependent on the makeup of your community.

in reply to Ariadne Conill 🐰:therian:

in the case of treehouse, we have 14 active moderators for a community of 560 active users.

this gives us a ratio of about 1 to 40. at one point in time, we had a ratio of about 1 to 15. something around this ballpark is probably the golden ratio.

if you have automation to deal with trivial abuse such as crapflooding, then you need less moderators.

in reply to Ariadne Conill 🐰:therian:

it is also important to realize that moderation is effectively a form of "community policing".

this means that it will only work if the folks doing the moderation are respected members of the community.

this means that you shouldn't bring in outsiders to do your moderation -- they need social capital inside the community in order to be effective.

in reply to Ariadne Conill 🐰:therian:

moderation is not necessarily about whacking things with a big stick either.

most of the time, simply taking a user aside and saying "can we not?" is enough to get the behavior to stop.

freenode (RIP) referred to this approach as catalysing. libera.chat has an entire playbook on it: libera.chat/guides/catalyst

the goal of moderation is to keep productive conversation flowing and the temperature low.

⇧