This is the whole formula. It is published because a ranking whose rules nobody knows gets played in ways we did not intend, and we only find out once it has already happened. A ranking you can read gets played by doing more of the thing that carries the most weight, and that thing is the one you cannot buy.
The weights
Measured activity
Messages sent by people in this server over the last 30 days, counted by the bot. Log scaled, so a server ten times larger does not score ten times higher.
How hard to fakeHard. Faking this means real accounts having real conversations, which is the thing you were looking for anyway. This carries the most weight because it is the part nobody can buy.
40%
of the score
Bumps
How many different people ran the /bump command inside the server over the last 7 days. Each person counts once, however often they run it, and the most recent bump from an hour ago counts double one from a day ago.
How hard to fakeBumping is self-service, so counting bumps would only measure how patient one admin is. Counting people instead means an owner who bumps every two hours all week is worth the same as an owner who bumped once, and the only way up is other members choosing to do it too.
20%
of the score
Votes
Votes from logged-in Discord accounts over the last 30 days, weighted by how old the account is and frozen at the moment of voting.
How hard to fakeThe easiest part to farm, so it is deliberately not the largest. One vote per account per 12 hours, new accounts count for less, and once more than 3 different accounts vote from one address on one listing within a day, the extra votes stop counting toward the ranking.
15%
of the score
Member reviews
Ratings from people the bot confirmed are members, and have been for a few days. Counts from 5 reviews upward; below that it scores zero.
How hard to fakeThe most expensive signal to fake that we have. An owner can ask their own members for reviews, and so can anyone anywhere. What they cannot do is have non-members write them, and every list without a bot in the server allows exactly that.
15%
of the score
Retention
How many of the last 30 days of arrivals are still around, from joins against leaves. Too few arrivals to say anything means it does not count at all.
How hard to fakeA server that keeps people is better than one that funnels them through. Low weight, because letting nobody in would otherwise look like perfect retention.
10%
of the score
What is deliberately not in it
Member count
It is the one figure an owner can buy outright, and the oldest trick on every server list. It is shown on the card and used for the size filter. It never moves the order.
Anything staff can turn up
Featuring and boosting a listing float it to the top of the list, visibly, as an editorial decision with a person behind it. They are no longer part of the score, because a score staff can turn is not a measurement.
Money
There is no paid position in this list. The price of a listing is that the bot is in your server, which is also what makes the numbers on it possible.
When something was not measured
A window counts as measured once at least 7 of the last 30 days carry data. Below that, activity and retention drop out of the formula entirely and the remaining parts are rescaled to fill it. A server that was added last week, or where the bot cannot read the channels people talk in, is judged on what is known about it rather than pushed to the bottom as if it were dead.
Nothing gets filled in with an average. A missing measurement is shown as missing: the card says it is not measured yet, and so does the strip on the profile. That is the cost of having no numbers here, and it is a visible one rather than a hidden penalty.
Measured and quiet is not the same thing. A server with a full window of data and nothing said in it scores zero on activity, because that is a finding.
What is never published
Anything about an individual member. These pages carry totals for a community and reviews people wrote under their own name, and nothing else.
Counts of blocked scams or moderation actions. A high number can mean a good filter or a server under attack, and publishing it invites people to attack themselves for the badge.
The exact protection level a server runs. Protected yes or no is public; the setting behind it would tell an attacker precisely what they wanted to know.
Scores are recalculated every 15 minutes. Found a listing that does not add up? Report it from its page and staff will look at it.
