How a drill runs
Seven steps: read your filters, drill your mods, let Twitch grade the filters, find rules that won’t ban regulars, prove them on attacks they’ve never seen, then push with undo.
Step 1
Read your filters
ModScrim reads your channel’s public blocked terms and AutoMod levels from Twitch, and starts a 7-day chat log used for one thing: checking that new rules won’t ban your regulars.
Twitch doesn’t share private blocked terms with apps, so ModScrim can only see and test the public ones. It never changes your AutoMod settings.
Public blocked terms
free nitrobit.lybuy followersfreebits- Hostility
- Discrimination and slurs
- Chat log
- 1,284 messages
Step 2
Set up the drill
Pick tactics, the number of attack and banter lines, the speed and the slip window. Share the mod link and the 6-digit join code, and add the overlay to OBS as a browser source.
Rehearsing alone? Turn on Kai (sim), a simulated co-mod who is always labelled and scored separately.
- Tactics
- Giveaway links, viewer spam, promo bots
- Lines
- 14 attacks, 8 banter
- Speed
- a line every 1.5 s
- Slip window
- 9 s
- Join code
- 482 913
Step 3
Run the drill
The practice chat fills with attacks written to slip past your filters, mixed with harmless banter. Mods delete, time out or ban from any device, with the keyboard or with buttons. The first action wins, and a line nobody touches within the slip window slips through.
Viewers see the overlay: the alarm, the scoreboard and a practice chat whose message text is blurred into bars.
nitrodrops_93fr33 nitro for the first 50 viewers → nitrodrop.invalidBanned by you · 2.1 sgrandpa_gamerLOL the jump at the endFalse ban: that was a regulargrow_fast_77want more v i e w e r s? best promo viewbump .invalidSlipped throughgiftsnow_botskins giveaway is live, claim yours at giftsnow.invalidSam got it firstStep 4
Twitch grades your filters
Every attack line goes to Twitch’s Check AutoMod Status, which says whether your current AutoMod settings and blocked terms would let it through. You get one plain number and a line-by-line table with the tactic and trick each line used.
Twitch limits how many checks a channel can make each hour. When the limit runs out, ModScrim uses its own matcher and labels every number it produced “local estimate”.
9 of 14
attack lines would get past your filters
graded by Twitch local estimate when the limit runs out
Step 5
Find better rules
The rule finder proposes up to six blocked terms in Twitch’s own syntax that catch what got past. Each one is checked against the last 7 days of your real chat and a pool of harmless banter, and any term that would match a regular is rejected, with the line it would have hit.
Some tricks can’t be expressed as blocked terms: spaced or dotted letters, zero-width characters, emoji inside a word and look-alike letters. ModScrim says “needs a bot rule” for those instead of pretending.
- Catches 2, 0 real-chat matches
giftsnow*Proposed - Would catch 2, but also matches “dm me the build later”
dm meRejected - Spaced letters, 1 lineneeds a bot rule
Step 6
Prove it on a fresh wave
Before anything is pushed, ModScrim builds a second wave of 30 attacks from a new seed, using tricks the rule finder never saw. You see how many were blocked before and after, what still gets through, and how many real messages the new rules would match. Push stays disabled unless that number is zero.
On a real channel, Twitch re-grades the fresh wave after the push, and the report shows how often Twitch and the local matcher agreed.
New rules matched 0 of 1,284 real and harmless messages.
Step 7
Approve, push and undo
Press Approve and push, and ModScrim adds each term to your channel’s blocked terms on Twitch, with a result for every term. Undo this push removes exactly the terms that drill added. Both are written to your audit log.
The after-action report keeps per-mod scores, what slipped, the Twitch grade, the proposed, rejected and pushed terms, and the fresh-wave result. Share links carry numbers only, never attack text.
Recorded in your audit log with the time and who pressed it.
What ModScrim can’t do
Stated up front, so the numbers mean what they say.
Some tricks need a bot
Blocked terms can’t express spaced or dotted letters, zero-width characters, emoji inside a word or look-alike letters. Those are grouped under “needs a bot rule” for a tool like Sery_bot, Fossabot or Nightbot.
Two engines can disagree
Twitch’s check also applies AutoMod’s categories and your permitted terms, which ModScrim’s local matcher doesn’t model. Every number names the engine that produced it.
Only public terms are visible
Twitch doesn’t return private blocked terms to apps, so the rule finder can’t take them into account.
It isn’t a moderation bot
ModScrim never acts on your real chat. Keep your AutoMod and your bots; ModScrim helps you tune them and train the people around them.
