We identified the source of the issue and rolled back the responsible change at 12:06 PM PT (19:06 UTC). Error rates have returned to normal, and we are actively monitoring to confirm full stability.
There is no data loss. No customer action is needed at this time.