SFT.CATCH // layers
What Sift catches.
In-depth guides for each type of flag — instant rules that never use a smart check, categories you can tune in /settings, and checks that always stay on.
How layers stack
Instant rules run first (blocklist, slurs, scam hosts, floods). Smart checks read the message and recent context. Child safety and self-harm always run on scored messages. See the main docs for setup and commands, or Running Sift guides for workflow deep dives.
Instant rules
No smart check used — still run when monthly checks run out.
- Block + strikeCustom blocklistPhrases your mods add in /settings — matched on word boundaries with normalization for spaced-out and leetspeak spelling.
- Block + strikeSlur fast pathKnown slur patterns are removed instantly, before a smart check.
- Block + strikeScam and lookalike linksTyposquat domains and known bait hosts are blocked without using a smart check.
- Block + strikeFloods, repeats, and mass mentionsRate limits and pattern rules for fast posting, duplicate text, @everyone spam, and the same post in many channels.
- ReviewOther servers' invite linksInvites to Discord servers you do not own are sent to review; your own server's invites are allowed.
Smart check
Reads the message and recent context; you can turn categories off in /settings.
- Block at default thresholdHarassment and personal attacksBullying, demeaning someone, or targeting a person — with context from recent messages when it matters.
- Block at default thresholdHate speechSlurs and contempt toward protected groups, including phrasing the instant slur list might miss.
- Review by defaultSexual content (adults)Explicit sexual content between adults — usually flagged for review first, not instant deletion.
- Block at default thresholdViolence and threatsThreats or glorification of violence against people.
- Review by defaultSpam, ads, and scams (smart check)Unsolicited ads, scams, and phishing wording — including leetspeak and spacing tricks the instant rules might miss.
- Review; urgent page when severeDoxxing and private infoSharing someone else's phone, address, or private contact info without consent — with context so sharing your own number is not treated as doxxing.
- Block at default thresholdIllegal goods and servicesSelling or facilitating drugs, weapons, stolen goods, counterfeits, and similar.
- Block at default thresholdMalicious links (smart check)Phishing and malware promotion when the link or wording is suspicious but not on the instant host list.
- Review by defaultImpersonationPretending to be staff, another user, or an official account to deceive people.
- Block at default thresholdExtremism and terrorismPromotion, glorification, or recruitment for terrorism or hate groups.
- VariesFilter evasionSpaced-out letters, lookalike characters, and words split across messages — joined before scoring.
Always on
Not in category toggles; separate thresholds.
- Block; urgent page when severeChild safetyAlways evaluated — grooming, exploitation, or sexual content involving minors. Cannot be turned off in /settings.
- Support DM; page mods when urgentSelf-harm and suicide riskAlways evaluated — from passive hopelessness to immediate danger, with hyperbole filtering and private support DMs.
More
Max rules, member reports, names, and server protection.
- Block or review (your choice)Custom rules (Sift Max)Up to ten plain-English rules per server — block or review when broken.
- Case for modsMember reports/report and Report message — human queue items, not automatic bans.
- Block or warnNames and thread titlesNicknames, display names, and thread titles — blocklist, slurs, and staff-impersonation patterns.
- Alert / stop abuseRaids, mass deletes, and anti-nukeJoin bursts, channel wipes, and mod-action abuse — alerts and automatic lockdown or rollback.