Registration screening for Fediverse approval queues. Bots rejected at the edge, humans approved in seconds — and the ambiguous cases left where they belong: with your moderators.
41 seconds of the live service: a person applies — approved. Twelve signups arrive from one subnet with identical text — flagged, then blocked as the pattern builds. Another person applies — approved, untouched. Real service, real verdicts, nothing staged.
Open registrations invite bot farms; closed registrations kill communities. Most instances resolve this by making moderators read every application — which works until the day a farm files fifty. Sentinel Signup screens the queue automatically so moderators only see the cases that genuinely need human judgment.
Applications are scored on behavior, never on newness. Being a new account is not a signal; acting like a coordinated batch is:
Reads human. Approved without a moderator touching it.
Ambiguous. Held for human review — the moderator decides, never the machine.
Coordinated pattern established. Rejected at the edge.
Every claim below maps to a dated behavioral proof. Entries marked pending are being re-run and captured for publication — a claim never ships as verified without its log.
Runs as a service in front of your instance’s approval queue — your moderators keep their existing workflow and simply stop seeing the noise. Free during beta.
Candor works at the intersection of gaming safety, age assurance, privacy, parental consent, cryptographic evidence, and auditable trust & safety decisions. Related: