Durable hub-to-hub sync — A persisted per-peer mirror (peer_journal) that survives restarts, per-peer due times after success and failure, bounded concurrent drains, exponential backoff isolation, and an atomic status snapshot after every cycle. One slow peer cannot serially block every healthy peer.
Fail-closed journal integrity — The daemon holds one kernel-backed writer lock, validates a complete page before I/O, fsyncs the active journal before replacing its owner-only cursor, rotates bounded immutable checksum-chained segments, and refuses startup unless the strict segmented tail equals that cursor. A failed poll never advances the cursor.
Live trust reload — Changed mTLS and Ed25519 trust material is validated as one immutable generation before any socket opens.
Per-peer operations — Strict versioned configuration, separate credentials and cadence, protocol negotiation, status snapshots, and a read-only doctor pass.
Offline licence admission — Named-license documents are verified locally before credentials are read or network connections begin. No account lookup or telemetry.
Lease-correct forwarding — Remote claims route only through the configured owner; denials and malformed or expired verdicts are never cached.
Signed audit equality — Independently signed commitments let operators verify equal Core log state offline across deployment boundaries.
Operator doctor pass
$ synapse-fleet doctor READ ONLY
hub-a → hub-b    caught_up
mirror integrity  verified
trust generation  current
remote claims     observed only