This page intentionally stays high-level. The exact signals, weights, and thresholds Sentinel
uses aren’t published, because doing so would make it easier to deliberately stay under the
radar. What follows is enough to understand what Sentinel does and how to work with it — not
how to game it.
Deterministic, explainable scoring
Sentinel does not use an opaque machine-learning model. Every score it produces is built from a list of specific, named reasons — things like account age, link reputation, or unusual message velocity — and that list is shown to your moderators alongside the score itself. When Sentinel flags something, the explanation is the score; there’s no black box to trust blindly.Raid protection
Sentinel watches join activity for patterns that look like a coordinated raid rather than normal server growth, and looks for corroborating signals before escalating. Containment increases in stages — from heightened verification, to quarantining new joins, to restricting messaging, up to a full lockdown — rather than jumping straight to the most disruptive response. Containment never quietly stands down on its own; a moderator releases it once things are safe.Anti-nuke
Sentinel tracks destructive or high-impact actions per staff member or compromised account — things like mass channel deletion, unauthorized permission grants, or unexpected webhook creation — against limits your guild controls. Depending on your configured enforcement mode, a violation can range from an alert to automatically quarantining the actor and stripping dangerous roles. The guild owner and AEGIS itself are always exempt, and AEGIS never attempts a role change Discord itself wouldn’t allow.Enforcement modes
Every protection Sentinel powers — raid, anti-nuke, message and link protection — has its own enforcement mode, set per guild:off, dry_run, log_only, human_approval, or
automatic. New protections start cautious (logging or requiring approval) rather than
enforcing immediately, so your team can see what Sentinel would have done before letting it act
on its own. See Quickstart for how to review and change these.