here's exactly what happens. openai's moderation endpoint is one check, not storage. a line goes to it only after passing my own word filters, and a screen name gets checked once. the call comes from my server, so they see my server, not your browser or address. their docs list that endpoint with no retention and no training use. i can't audit their side, but that's the published policy.

on my end, posted lines stay in the room's history with how long ago they were said, until newer lines push them out. my filters get patched as edge cases show up, so i keep a small sample of refused lines for 48 hours to catch any that refuse normal chat.

Ty good catch, will fix on the next update!

Fair. I leaned on AI, for the code and for picking and drafting a lot of the recent additions. The comments in the source were its working notes; I've stripped them from the build. I overstated "by hand" in the post: the rules and the final say are mine (no algorithm, no tracking).

this is really useful, thank you. "start from zero each time" is the gap for sure. a camp you build up from what you bring home is next on my list :)

thanks, really appreciate you playing. that's fair, right now each run stands alone apart from the seasons. what would keep you coming back, unlocks, a longer campaign, something to build at camp?

Thank you :-) - yep, it's me. no account or email. your lines are stored with your screen name and shown to people in the room for up to seven days.

a random token in your browser holds your name so nobody else can take it; it isn't tied to who you are. your IP is used for rate limits and spam and kept 7 days at most.

lines are checked by an outside moderation service before they post, and a few refused ones are kept 48 hours to tune the filter.

the bot check (Vercel) may set its own cookie. nothing is sold. full privacy page is in Help. don't put personal info in the room, anyone can read it.