booper learns from what people send it, which makes the corpus a shared space. These rules exist to keep it one worth learning from — and to keep everyone, bot included, on the right side of Discord's rules.
Don't feed booper hate speech, slurs, harassment, threats, sexual content involving minors, content that sexualises anyone without their consent, instructions for causing harm, or anything illegal. A content filter already refuses to store or repeat a lot of this, but the filter is a backstop, not permission to probe its edges. Deliberately trying to make the model say awful things later is the one unforgivable move here.
Only opt yourself in. Don't paste in someone else's messages, private conversations, or personal information — names tied to addresses, phone numbers, or anything that identifies a real person who didn't consent. The whole design keeps identities out of the dataset; don't smuggle them back in through the text itself.
No using the bot to harass people, evade server moderation, spam channels, or launder content a server has banned. Everything in the Discord Community Guidelines applies to what you do with booper, including what you make it say.
No flooding the bot with automated traffic, scripted mass corrections, or coordinated attempts to skew what it learns. The corpus is meant to be real people talking; keep it that.
Contributions that break these rules get removed from the corpus and excluded from future training and publishes. Repeat or serious offenders get ignored by the bot entirely, and where Discord's rules require it, reported through Discord's own channels. We'd rather never do any of that — the rules above are genuinely all it takes.
If booper says something harmful, or you spot corpus abuse, tell us: github.com/frgmt0/babble. The blocklist and filter improve fastest when people point at what slipped through.