Pulled the alert: `api-gateway p99 > 800ms for 8m`. Cross-checked last 30m of logs, deploy history, and the canary metrics. Cause: a Redis connection pool exhaustion after the 02:47 deploy. Not a real customer-impact incident — synthetic traffic from a load test. Silencing the alert until 06:00 and opening a ticket for the pool config. Going back to sleep — page me only if it's real.