Retry storms that look like sync outages
A silent backoff loop can look like a regional failure when you only read the conflict spike.
On sync endpoints, a client that retries without jitter can multiply conflict volume for a single user action. The board shows a cliff; the product never asked those users to leave—or to stay.
Before rewriting the sync path, walk the journey while watching live lag and error rates. Note client-only retries versus server confirmations. Check whether staging shares production keys.
Conflict Signal Mapping stays narrow on purpose: one sample window, deep reconciliation, a checklist you can rerun after every release so the cliff does not return unnoticed.