max / makenotwork

multithreaded: implement the four livechat traits MT's side of the crate seam: ChatRooms, ChatAuthz, ChatIdentity and ChatModeration, plus the process state they hang off. Still no routes. ChatAuthz is the one that mattered. src/routes/scope.rs records the C1 chronic (three ultra-fuzz runs of handlers evaluating predicates against a community the slug did not name), and chat is a new surface on those same URLs. Two structural properties keep it from being the fourth instance. A gate holds the Room resolved from the slug, every predicate reads room.id, and no method takes a community id, so the divergence cannot be expressed. And good standing is now evaluate_write_access, one function that check_write_access renders as a 403 and chat renders as a DenyReason: two renderings, one predicate set, so a check added for the forum reaches chat and the two cannot disagree about who is muted. The gate is per-request because fan_plus reads UserPerks, which lives on the session rather than in the database. It refuses outright when asked about a user it was not built for, rather than answering with the wrong person's perks. Chat moderation reuses the existing stack rather than adding one. A timeout is a mute with an expires_at, a ban is the same community_bans row a forum ban writes, and every action lands in mod_log on the same transaction as the mutation it records, so a purge cannot land without its audit row. Four ModAction variants added; the message id travels in the reason text because target_id is a UUID column and a chat message id is a BIGINT. Room state folds policy, community state and suspension fail-closed: off and suspended close the room, Frozen and Archived go read-only via allows_writes_for_members() rather than a variant match, so a new CommunityState gets the same answer chat and the forum both give. Identity is cached with a TTL and a hard cap, invalidated on rename. SessionUser carries no avatar_url, and the uncached shape is one query per author per reconnecting client at the exact moment every client in the room reconnects at once. Hub and rate limits are stated explicitly against the 512M cgroup cap rather than inherited from the crate defaults, and chat gets its own send budget: USER_POST_RATE_LIMIT (15/60s) governs forum posts and would throttle a normal conversation. 25 trait tests including the full four-policy read/write matrix; 336 integration and 192 unit tests pass, clippy and fmt clean.