Detect coordinated join waves and stop them intelligently.
A raid is never one member but a group arriving together. The system links joiners to each other through shared fingerprints and gives every group a suspicion score. A suspicious group is not punished for arriving; its first messages decide. Hostile behavior confirms the raid for the whole group, while quiet groups clear automatically.
Trusted roles and trusted members are never placed in a suspicious group. Returning members and long-lived accounts are hard to flag by design.
A raid is never one member but a group arriving together and executing one plan. That is why judging each joiner alone does not work. The system links joiners to each other through seven shared fingerprints: a shared server tag, the same invite link, accounts created together, matching usernames, bare profiles with no avatar, account age, and shared other servers. The more independent signals a group shares, the higher it scores.
The smartest part is that a suspicious group is not punished for arriving. Intent analysis watches its first messages: many mentions in one message, members posting the same text, or a burst of messages in seconds. Hostile behavior confirms the raid for the whole group, while a quiet group clears automatically. So a group of friends joining together is never wronged.
You can also enable an AI second opinion on borderline groups: it reads the evidence from ages, tags, and patterns, and confirms a raid or acquits the group. It never acts alone below high confidence.
When a raid is confirmed, raid mode arms automatically: suspicious joiners are contained on sight while normal members keep joining freely, alongside paused invites and a raised verification level. Everything reverts when the duration ends. The page shows the live status and the detected groups with their scores, plus buttons to ban or pardon a group by hand.