How your data is used

This site exists to collect real, consented examples of toxic/abusive/sexual/ threatening/hateful language to train Profanix, a content-moderation model built to catch exactly the kind of language you're trying to sneak past it here.

What we store

Every submission is stored with two separate consent flags you set yourself:

• Training use (required to submit at all): the text and the model's scores for it may be used to retrain/evaluate Profanix.
• Public display (optional, off by default unless you check it): your score may appear on the public leaderboard next to your chosen display name. We never publicly show the raw text you submitted, regardless of this setting — only your score and which category you beat the model on.

What we don't allow

Don't target real, identifiable people — write about fictional or generic targets only. Submissions that look like they contain a real handle, email, phone number, or link automatically lose public-display eligibility even if you checked that box, though they can still be stored for training with your consent.

Identity

No account or email is required. We set an anonymous, random cookie on your first visit so we can attribute submissions to a leaderboard handle you pick — that's the only identity this site keeps.

Model version

Live scoring here uses Profanix's small baseline model (char n-gram TF-IDF + logistic regression), not its larger transformer/hybrid variants — it's fast enough to score every submission instantly and is what this game is specifically trying to find weaknesses in.