Publishing
The publishing reliability system
Claims prevent double-posting, retries absorb transient errors, and stale-claim detection surfaces interrupted publishes for manual recovery.
Last updated June 14, 2026
Publishing to third-party platforms is unreliable by nature: networks time out, APIs rate-limit, tokens expire. Signal's reliability system is built so those realities never turn into a double-post or a silently lost post.
Three mechanisms work together: an atomic claim before each publish, a retry policy with backoff for transient failures, and stale-claim detection for the rare case where a publish is interrupted mid-flight.
The guarantees
- ·No double-publish — an item is claimed (scheduled → running) with a guarded update; only one tick can win the claim.
- ·Transient errors retry — a timeout or 5xx is retried with backoff, not failed immediately.
- ·Interruptions are visible — if a tick dies mid-publish, the item is left as a stale claim and surfaced for manual recovery rather than blindly re-published.