There is a rule in the design of software and user interfaces called the Principle of Least Surprise. It says that when a person interacts with your software or interface, and they take an action such as filling out a form or clicking a button, the correct outcome is whichever outcome is unsurprising, or, if all possible outcomes would be surprising, the one that surprises them least.

This is not an ironclad law. And there are subtleties. For example, if a bank website has a button to 'Close Out Your Account', in some sense the least surprising result would be to FedEx you, the user, a large quantity of $100 bills, and this is generally speaking a bad idea they should not do even if you request it of them. (Also probably illegal.) However, the principle of least surprise says that their error is not 'doing that dumb thing,' nor is it refusing to do it; their error is having that button at all. Try instead 'Close Your Account & Transfer' or 'Begin Closing Your Account.' Either of those will produce less surprise when, instead of the dumb thing, they present you with a form asking for information. The bank wants to tell you that a money transfer elsewhere is the option they are offering, and so ensure you will be unsurprised when they ask you to tell them an account number and routing number they will wire your money to.

There is a very similar principle at work for community moderation. At least, community moderation as I see it.

I approach community moderation with the perspective that it is the moderator's job to make things easy for its members, not the other way around. The point of a community is to provide value to those within it; the moderator is doing a public service by facilitating it, not serving their own interests. By a quirk of how communities tend to develop, the norm is for moderators and community organizers to be volunteers, not paid, but it's a part-time job, and often a difficult one, and needs to be approached as a job. If you cannot moderate well, it is often the correct decision to let the group die or fundamentally reduce the scope so that you can do it right, rather than let it continue as it is and do it wrong. Bad communities can and do make their member's lives worse, and it is irresponsible to facilitate a community that is or will become one. If you don't agree with this philosophy, you may find some of this advice inapplicable, and while I can try to convince you of it, that's outside the scope of this post.

Particularly, this is not a principle that works for lazy moderators who want a community but would rather have nothing than have one that took a lot of upkeep. I'm unsure whether it works for large communities, groups in the thousands; it doesn't scale very well as you need to add extra moderators, and moderator count scales pretty linearly with member count. Certainly it doesn't work for communities of millions such as /r/AskHistorians. Though I will assert that nothing else does either; communities that big are ungovernable. Also, this principle tends to be more important for 'nerds' who tend borderline autistic (or not so borderline), whose social skills are lacking or tuned for unusual subcultures, and who tend to be socially anxious at higher rates than the general population, as well as for other groups drawn from multiple cultures with different interpretations of 'normal.' This means it tends to be most appropriate for online communities like forums, chat servers, and comment sections, though it applies in many other places.

Certainly the easiest place to see this at work and test its results is online, because a forum or chat server has visible history. You can look back and see who said and did what, and deduce what was surprising to them and what was expected. It also allows you to take time, when making your decisions; most online communities are somewhat asynchronous, and so a moderator can generally pause to consider the right action without hugely disrupting things, in a way a community organizer in person generally cannot. Applying this offline has to come with caveats. But both online and offline, there is a simple, analogous rule anyway, and I would describe it as this:

If a major moderation decision comes as a surprise to the people affected, the moderation team has made a mistake.

Now, to be clear on things I do not mean.

I do not mean that you can't ban someone if they're surprised by the ban. "I made a mistake, but you're still banned." should not be your first resort, but it's on the table.

I do not mean that someone legitimately surprised by your moderation actions has acted correctly and ethically, though this is of course sometimes true.

I do not mean that this is the only thing required of your rules for good moderation. "I will ban anyone whenever I feel like it if they annoy me," if regularly, visibly used, will rarely be surprising. But that's not a very good policy.

I do not mean every person affected must have a near-perfect understanding of the rules for it to be ethical to enforce them. In many cases the best you can do is to make clear to someone that they are on notice and have to either cooperate with you to improve or be kicked out on the next offense. Frequently that next offense will not be something they believed to be forbidden or even borderline, but they will still be unsurprised in a larger sense, if you have warned them appropriately.

I do not even mean you have to take the self-professed surprise of people affected at face value. Though you generally should, because the human capacity to interpret people who are causing you annoyance and extra work as malicious is nearly infinite. Which makes it very tempting, and therefore dangerous, to assume dishonesty. (Even after you take this rule into account. Thank you, Douglas Hofstadter.)

Much like the example of a FedEx envelope full of cash, this does not go infinitely far. There are things you should absolutely not do when moderating a community, like keep around someone extremely disruptive who makes others uncomfortable and contributes very little to whatever interactions and goals you and your community value.

But if removing such a person comes as a surprise to them, you shouldn't have had that button.

Remove them, by all means, but then you need to look at your decision-making process and look for where the mistake(s) lived. Call this a postmortem, a retrospective, a failure analysis, an incident report; whatever it is, you should look at your rules, the warnings you gave, and any norms you established other than the explicit rules, and determine what changes could have prevented it. And whether those changes would be worth the cost, but generally they will be; predictability is worth pretty high costs.

This has some important corollaries:

Complex codes of rules are best avoided, because they are more difficult to understand and therefore more likely to surprise people affected.
In situations of edge cases in which a person is repeatedly coming close to the edge of what's permitted, it is still the moderator's responsibility to manage the situation, not the person in question. (Unless it's deliberate. As above, it's dangerous to assume it is deliberate.)
To understand the process, it is easiest to see it in action. Transparency to the community is therefore useful pragmatically and instrumentally, even if you don't terminally value it as a virtue in itself.
Ad hoc rules for specific individual instances of bad behavior should be considered 'technical debt.' They're difficult for other members to generalize, making your future responsibility to explain rules larger.

I have, as it happens, been a community organizer more in person than online. Despite that, I'm more confident in applying it online, because there it is much easier to learn from someone else's mistakes by reading history. From that offline experience, though, I have some notes on applying the principle in person.

Firstly, I don't recommend trying to learn it in person, if avoidable. That will be significantly harder, with no ability to record information (especially since much more comes in tone and posture and timing) and much reduced ability to think over your decision before it's needed. Once you do, that same extra information should make it easier to put into practice, but it's harder to get there.

Second, you're in a bind in the middle ground, when you're pretty good at thinking through the least surprising decision, but not up to the 'snap judgment' level. So what you want is to set your policy in advance. But this is less flexible and inherently significantly more exploitable. Which makes the practical problems in implementing it with good intentions jump significantly.

So, third, two practical rules of thumb I would recommend if you want to try this and mostly run in-person real-time events:

Don't make permanent decisions in real time.
Prize transparency as highly as you can afford to.

An in-person community is a repeated interaction, and kicking someone out of one event and following up with a permanent decision later is feasible and usually a good tradeoff for a complicated situation. And then for those permanent decisions, and for many other decisions, the lack of history to inspect makes the verdicts obscure and hard to understand, leading to uncertainty, anxiety, and surprise; therefore, transparency about what has and hasn't merited bans or other consequences and who is and isn't 'on probation' is unusually valuable.^[1]Online you can let people ask, but friction is always somewhat high, and making offline friction low enough to make it available in practice is essentially impossible.

In general, what this principle and predictability provide your community's members is stability and reassurance. If expectations are clear, and there is an understanding that the moderators care about being predictable, community members can largely not worry about whether they're coloring inside the lines. This is why it's particularly appropriate for communities with imperfect grasp of, generally nonstandard, or just widely-varying social norms, and/or high rates of anxiety; those are communities where the fear of transgressing unexpectedly, whether legitimate or not, is high. Even outside those circumstances, though, you want people to be spending less time checking whether what they're doing is technically allowed, and more time doing whatever it was they joined to do. Perfect rules, and perfect communication of them, are impossible, even when everyone has the same basic assumptions; this principle is a good way to ensure that your deviations from perfection are reliably in the direction the people of your community don't need to worry about.

If you can't get perfection, get harm reduction.

This is harder than it sounds, and probably doesn't sound easy. There are serious tradeoffs that make full transparency fraught, if not outright dangerous due to libel law. 'As highly as you can afford to' may not be that high. But it's worth trying, and keeping in mind. ↩︎