What the Last Major Outage Taught About Feed Failover

The most recent major cloud outage (regional networking, affecting market-data paths) replayed the classic failure sequence:

• Upstream stalls — the feed goes silent; clients see stale data without knowing it's stale
• Timeout + reconnect — clients disconnect, hammer the failover endpoint simultaneously
• The double failure — failover endpoint wasn't load-tested for 100% traffic; it buckles

The design rules:

1. Stale-data detection — if no tick in N seconds, mark the feed stale; a gray "delayed" badge beats silent wrongness
2. Regional failover — tested at full load, not just "can it start"
3. Graceful degradation — show the last known good quote with a timestamp, don't show nothing

Stock Picks's quote layer degrades this way — users see "delayed 2m" instead of a frozen screen.