A regional load balancer intermittently returned 503 for approximately 26 minutes affecting the dashboard only. API ingestion was unaffected. Root cause was a bad config push to Cloudflare's L7 rules; reverted. Post-mortem published Sep 29.

Planned Postgres failover test. 8 minute impact to the US-based dashboard replica. Zero data loss. Notified in advance to Enterprise accounts.

Logs from GKE Autopilot clusters were delayed by up to 15 minutes due to a change in GCP Pub/Sub's quota behavior. Caught up within 83 minutes. Monitoring improved to catch similar lag within 5 minutes.