• Adults-only demo built in a sandbox — not a real service.
• Accounts ride on your Claude sign-in — no password to invent or leak. Your @handle is unique and your profile follows you across devices.
• The compatibility score comes from a plain 10-item Big-Five-style test at signup. What cards show is a pairwise match % against your answers — there is no global ranking. The vector (and answers, for cross-device editing) lives in the shared database like the rest of the profile.
• Demo accounts (🧪) skip photo verification with generated placeholder cards — labeled everywhere, never shown as verified, resettable from this tab.
• The 3-face-photo rule is checked by Claude vision (a face in each, same person across all three). It's advisory-grade, not biometric ID; profiles that skip or fail it are labeled unverified and filterable.
• Your photos live in this page's shared database like everything else — people who can open this link can see them. Ghost mode hides your card (photos included) while it's on.
• Ghost mode: you vanish from decks and house counts, appear in rooms as an anonymous 👻, and your messages ride a ghosts-only channel that non-ghosts never see.
• Profiles, likes, matches, houses and DMs are stored in this page's shared database: people who can open this link can see them, and two real people here can genuinely match and chat live.
• Spin partners are AI personas (always labeled). They're not humans.
• Distances on cards are simulated — the app never reads your location.
• There's no real call infrastructure, so there are no device-to-device video calls. Your camera stays on your device — except the optional low-res "live snaps" you can share with people in a house.
• The Ranks boards are strictly opt-in. Opting in publishes your name, avatar, like/conversation counts and live-chat time to everyone on this link; opting out deletes your row. Floors (5+ likes, 2+ chats) and a 30-min session cap blunt the obvious ways to game them.
• Friends, maybe Benefits: your 🤝/🍒 consent toggle broadcasts to everyone in the session (tiles and spins show it), defaults to 🤝 every session, and the AI personas obey it.
• Reporting (FmB) attaches a rolling client-side buffer of the last ~60s (chat + any live snaps). Nothing is uploaded or stored unless a report is filed — a clean session leaves zero footprint. Filed reports are visible to both parties under Me → Safety (and, prototype caveat: to link-holders, like all shared data).
• After Dark runs on 👍/👎 peer ratings (rate a spin partner or tap a tile). ${"5"}+ ratings with over 60% 👎 bars an account from After Dark; standing shows under Me → Safety. Ratings are one-per-rater-per-person and forge-resistant.
• 🧲 Magnet Boost (rewards): top-N on Convo rate or Stamina (default 3) claims one token per board per ISO week; deploying pins your card first (with a visible 🧲 chip) for a window (default 30 min) in every deck where show-me aligns both ways, bypassing prior passes and all discovery filters except bans/bars/photo rule. Already-matched or friended viewers are excluded; a pass during the window dismisses you for that window only — a later boost pops again. N and duration are operator-tunable.
• Unranked accounts carry daily caps (default 20 swipes incl. ghost deck, 5 spins; operator-tunable, meters shown, resets at local midnight). Opting into the 🏆 boards lifts both instantly; opting out re-imposes them. The intended effect is gravity toward the self-governing ranked ecosystem.
• The 📉 Losers board is compulsory: like→match conversion, ISO weeks (Mon 00:00–Sun end), ${"10"}+ likes to appear, worst first. Last week's bottom quartile drops to the slack ad tier (interstitials every 8th swipe + spin ads); the bottom 1% on 30+ likes under 5% conversion gets a 30-day swipe bar (Clubs and Spin stay open; barred cards leave everyone's decks). Floors exist so low-volume and new users can't be "reckless" by accident — selectivity is always the way out. Enforcement is client-side here, like the other rails.
• 👻 Ghost profiles: a second, photos-only identity for the anonymous Meet deck in ghost mode (no name/details, no face rule, 1–3 photos, editable from Me). Ghost likes/matches/threads are a parallel anonymous market; the ghost Consensus is a separate ledger ("g:" targets) rated from ghost threads. Ghost bans: avg < 2.5/10 over 15+ reviews → timed ghost lock; 15 zeros → permanent ghost ban. Real side untouched.
• 🛠 Operator control: the panel includes direct account actions (platform ban/unban, ghost ban/unban, lift swipe bar, delete profile) via the owner-write opbans rail, live for everyone. On the website build, the operator is an email+password Firebase account; its email is honored by the database rules themselves, and a sign-in button appears under Me.
• 🛠 Operator panel (artifact owner only, in Me): every ban/punishment/display threshold above is live-editable through a shared config doc, plus manual-review queues flagging extreme rating patterns (receivers of mass 0s/10s, and raters who only hand out extremes).
• Consensus is a 0–10 experience rating anyone can leave after meeting you (club tiles, spins, DMs, ghost threads). It's separate from match % (a test) and the AD 👍/👎 (AD access): a Market Rating appears to others only at 15+ reviews; 100+ raters with a majority of 1–2s closes the account; raters are anonymous and forge-resistant (one slot per rater per person). You never see your own number — in this prototype that's a UI rule (shared data is link-readable); a real deployment enforces it server-side and, in the EU, pairs auto-closures with a statement of reasons and appeal.
• Spin ads are tiered: 6m+ average live-chat time earns an ad-free Spin; averages under 2m or a 👎-heavy rating sees an ad after every spin; everyone else sees one occasionally. New accounts get a 3-chat grace window, sessions cap at 30 min, and the tier + reason shows above. Data used: on-device time counters (separate from the opt-in boards, never published) and the public 👍/👎 ratings. All ads here are fake demo/house ads.
• Enforcement binds to the signed-in account — no IP or geolocation harvesting here; real deployments get IPs at the server under legitimate-interest abuse logging, not via ToS mandates.
• Lock and moderation beyond the above are UI-level only here.