Appointment reminders at noon, and a catch-up pass at six

Each scenario reads a cached token, works out tomorrow's date and splits into one branch per clinic site, each fetching that site's appointments from the scheduling system and iterating them. A one-minute daily window makes each effectively a once-a-day run.

A filter drops calendar entries whose name is a placeholder (a block, a break, a site label) and a list of excluded practitioners, so reminders go only to real bookings.

Each reminder is an approved template with the patient's first name, the time and the practitioner. A row in a reminders log records the appointment id, times, practitioner, site, reminder type and send time.

The evening scenario runs the same pipeline but looks up each appointment id in the log and continues only if there is no row. It reaches the bookings made after midday and writes its own log row, so neither pass can repeat the other.

If a send fails because the contact has no open channel, an error route looks the patient up in the messaging system, creates the contact if missing and retries. If that also fails, a row goes to a manual-follow-up sheet (patient, phone, the full message and a click-to-chat link) and processing carries on, so staff can send it by hand.