We experienced an elevated rate of retries due to transient timeouts from an upstream LLM provider. The service remained available throughout, and all queries were completed successfully after retry. Our monitoring incorrectly flagged this as downtime. We are adjusting our thresholds to better distinguish between degraded upstream latency and actual outages.

Write service recovered.

Write service went down.