Design the service that ingests parcel-tracking updates from many shipping carriers, normalizes their differing status vocabularies into one canonical timeline, and serves customer-facing tracking plus proactive notifications.

Operating context. Dozens of carriers report progress differently: some push webhooks, some must be polled on a schedule, and each has its own status codes, formats, and reliability. The service maps every update onto a canonical shipment state (label created, in transit, out for delivery, delivered, exception) and maintains an ordered timeline per tracking number. Updates arrive late, out of order, and duplicated. Customers open a tracking page expecting the latest state, and the service pushes a notification on meaningful transitions. Scale: about 100 million active shipments and roughly 40000 tracking events/sec at peak.

Out of scope. Carrier rate-shopping and label purchase, the notification rendering/delivery channels themselves, returns logistics, and address validation. Assume carriers provide webhook and/or polling endpoints and a separate service delivers the notifications you enqueue.

What to produce. A high-level architecture covering: how heterogeneous carrier inputs (webhook and polled) are ingested uniformly, how raw updates are normalized to the canonical timeline, how out-of-order and duplicate updates are ordered and de-duplicated, how the customer read path serves current status at low latency, how meaningful transitions fan out to notifications, and how an unreliable carrier is handled (webhook stall to polling fallback). Sketch the components and trace a raw carrier update from ingestion to updated timeline and a fired notification.