Notification rules

A notification rule reads as one sentence: when this happens, tell these people, this way. Everything else on a rule — the throttle, the schedule, the escalation — exists to stop that sentence firing more often than anyone can act on. Rules live under Customization, separately from the notification feed itself.

Every rule carries the same six parts.

• Trigger event — one of the ten events listed below. A rule watches exactly one.
• Conditions — optional filters on the event’s own data. All conditions must match; within one condition any listed value matches.
• Recipients — by role, by named person, or resolved from the event itself.
• Channels — push, email, webhook, or any combination. Each is attempted independently.
• Throttle — a cooldown that suppresses repeats for the same thing.
• Schedule and escalation — when the rule is awake, and who to chase when nobody acknowledges.

The list is fixed, and the rules page groups it the same way the form does. Trip events: created, completed, failed, cancelled. Stop events: departed, failed. Fleet and GPS: route deviation, signal loss detected, vehicle breakdown. Delivery windows: window violation, raised when a stop’s estimated arrival falls outside the window it was promised for.

Channels fail independently. A dead webhook does not stop the push, and a recipient with no email address is skipped without failing the batch for everyone else. Each attempt is recorded whether or not the send succeeded, so the dispatch history is a record of intent rather than proof of receipt.

Three limits sit between an event and an inbox, and they apply in this order. First the rule’s cooldown, which is held per rule and per subject — a cooldown on one trip never silences a different trip. Then the active window: outside it the notification is queued rather than dropped, and released when the window next opens. Finally a per-person cap of 50 notifications an hour, which protects one unlucky dispatcher from a storm no single rule caused.

Escalation answers the question the rest of the rule cannot: what if nobody acts. Turn it on and give it a timeout, and any notification that has not been acknowledged by then is re-sent to a second set of recipients, marked as an escalation and raised to critical priority. Leave the escalation channels empty and it reuses the rule’s own.

Escalation happens once. After the second wave goes out the original is taken off the eligible list, so an unacknowledged notification cannot re-escalate on every sweep until somebody clears it.