Moderation.
The one rule is no lying. Once strangers arrive, that rule has to survive strategic lies — bad-faith actors, brigades, sock puppets, and the rest of the modern internet's standard furniture. Here is the honest plan, held to the same floor as the rest of the house.
The principle
The same rule that governs the curator governs every visitor: no lying. Sock-puppets, fabricated quotes, stolen identities, faked credentials, brigades organized to lie in concert — all break rule one, and all earn the same response: removed from the room, named on the rafter when the receipt is clean.
The current state — honestly
- Most of the site is static. There are no comment threads on this side of the launch — nothing to moderate, because there's nothing public to post to. The games are local-only, no account, no server-stored speech.
- The online multiplayer Worker exists (the
ohs-roomsworker), but is held until Google OAuth + counsel land — exactly because of this question. Rooms with accounts are rooms with moderation needs, and we don't ship the rooms before we ship the moderation answer. - The dispatch board, the rafters, the Reading Room, the lexicon, the wings — all curator-written. The curator IS the moderator of the curator. The rafter is where the curator's own slips are named, in public, dated.
How enforcement works when rooms open
- One reporting address — [email protected]. Subject: moderation. The address is read by a human (the curator, until co-laborers join — see /serve-with).
- Steward authority per room. Each room (a party game session, a debate session, a thread) has stewards with authority to remove. In small rooms that's the room's host; in larger ones, designated humans (named on the rafter, not anonymous).
- Removal for lying is the default response. No three-strikes for fabrication — fabrication breaks rule one once. Disagreement, bad takes, mistakes, typos, and devil's advocacy are not lying and are protected. See the One Lie clause for the distinction.
- Appeals exist. Anyone removed can write the same address and ask for the receipt. If the steward got it wrong, the call gets reversed and the record corrected (on the rafter, dated).
- Bans are recorded, not hidden. If a person is banned, the reason — the specific lie, the date — is summarized in a non-personal way on the rafter. We do not silently remove and we do not publish private data.
What we won't do
- No mass automated moderation by LLM. The whole site argues that scoring is human work; we don't outsource the scoring of speech to the very machine the site is built to test. (We may use the machine to flag for human review, never to act unilaterally.)
- No shadow-banning. If a comment is removed, it is removed visibly; the user is told why. Silent suppression is a kind of lying.
- No private data trading for enforcement. No identity verification beyond what's strictly required, and no sale or sharing of who-said-what.
- No "trust and safety" theater. If a thing is broken, we say it's broken. If a feature can't carry its own enforcement weight, that feature stays held.
What we don't know yet
Named honestly: scale-tested moderation playbooks for paid, real-name rooms have not been written. Counsel is on the curator's list (CLAUDE.md 0g binds it). Until the playbook exists and counsel has signed off, the locked features (real-name login, the corporate hook, the paid debate league) stay locked. The hold is the moderation answer until the moderation answer exists.
No dirt in the museum — and no leaks. Both pillars; one house.