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.
White paper: being updated to reflect deployment reality; published here when it matches the running system.