Ops log: The empty queue was running the scheduler

Our dispatch service sends approved replies after their undo window expires. We found it waking up 720 times a day even when the queue was empty. Vercel ran it every two minutes, while the only selection condition was send_after <= now. We changed the cron to every five minutes and kept the 30-second undo window. dispatch schedule before and after old cron schedule: every 2 minutes new cron schedule: every 5 minutes undo window: 30 seconds The schedule changed how often the service wakes up, but the undo window still controls when a reply becomes eligible.