The action ladder
Warn, hide content, end the call, cooldown, temporary suspension, ban. Each step has a written trigger, a statement of reasons and an appeal link.
Where humans come in
Appeals, uncertain cases and child-abuse hash matches go to a vendor's on-demand reviewers. Regulator requests and tribunal orders go to the founder. There is no moderation team to manage.
What it talks to
- API on Cloudflare Workers: One API for the whole product. It checks who the member is and what they may see, then calls the right module.
- Chat and rooms: Keeps live state for one-to-one and group chats, plan rooms and the voice speed-meet queue: who is in, whose turn, how long is left.
- Media pipeline: Takes every uploaded photo through quarantine, a child-abuse hash check and an AI image check before it can be seen, then resizes and publishes it.
- Voice rooms: Carries the audio for voice speed-meets: short rounds between two verified members.
- Consent ledger: An append-only record of every face reveal, album opening and video grant, and every revocation. Members can read their own.
- Discovery and recommendations: Finds hubs, people and open seats that fit a member, and excludes anyone who has been reported, pending review.
- Translation and scam flags: Translates chat for free and flags money requests, crypto pitches and pressure to move off the platform.
- Human review on demand: Vendor reviewers who handle appeals, uncertain cases and child-abuse hash matches, plus a case-management tool for reasons, appeals and transparency numbers.
- Email, SMS and push: Sends notifications, statements of reasons and trusted-contact messages when a member shares a plan.
Designed for
- Shadow mode for two to four weeks before any classifier acts
- Flirting in dating conversation is not harassment: false positives are the real risk
- Every action has a reason and an appeal link
Where it appears in a request
- A photo upload: The policy engine applies the written rules: publish, block, or send to a human if a hash matched or the case is uncertain.
- A report against a member: Classifiers score the messages; the policy engine picks a step on the action ladder: warn, hide, end a call, cooldown, suspension or ban. Every step comes with a reason.
- A voice speed-meet: Live voice moderation listens for abuse under a policy written for dating conversation. It ran in shadow mode first to measure false positives.