Different providers may assign different event identifiers to the same business change. Deduplicate delivery within each source, then use an explicit business-operation relationship where cross-source duplicate effects are possible.
Event identity and operation identity differ
An event identifier names a notification in a provider’s documented namespace. A logical operation identifies the business intent or authoritative change. Several events can describe that one operation, and superficially similar events can represent legitimate separate work.
For a proposed receiver, scope source deduplication by provider, account, and event identifier. For shared downstream effects, maintain a separate operation key based on trusted business references and effect type. Do not generate it from personal data or a vague similarity score alone.
Keep original source identities in the receipt ledger. Collapsing everything into one opaque label can prevent investigation of which service sent which evidence.
Test a fictional double notification path
A booking system emits an appointment-created event, while a CRM emits a related opportunity-update event. Both integrations are configured to send the same customer confirmation. Their event IDs differ, so source-level deduplication cannot prevent two messages.
The downstream notification policy should identify the authoritative booking and the intended confirmation effect. Both paths resolve to that effect identity, allowing one send under the defined concurrency mechanism. If they refer to different appointments, each may legitimately require a message.
| Event relationship | Required decision |
|---|---|
| Same provider event redelivered | Delivery duplicate |
| Different sources, same intended confirmation | Shared effect protection |
| Same customer, different appointment | Distinct legitimate operation |
| Similar payload, uncertain relationship | Reconcile rather than merge blindly |
| Concurrent cross-source arrivals | Atomic uniqueness at the effect boundary |
Assert the actual destination side-effect count. Two local receipts marked deduplicated do not prove that an earlier external send was avoided.
Avoid overaggressive matching
Time windows and payload hashes can support analysis but may suppress legitimate repeated work or miss semantically equivalent events with different formats. Prefer authoritative operation references and a documented mapping.
Cross-source relationships add maintenance and can remain unresolved when systems expose weak identifiers. Make uncertainty visible and choose a safe manual path for consequential effects. Do not promise universal exactly-once behaviour across services that lack the necessary contracts.
This deduplication design is proposed. The voice platform case study describes related integration experience. Read stable customer identifiers for the separate entity-mapping boundary.
Cross-provider deduplication FAQ
Can provider event IDs be compared directly?
Only within their documented namespace unless a trusted mapping establishes a cross-source relationship.
Should all events for one customer be merged?
No. A customer can have many legitimate operations and effects.
Where should duplicate-effect prevention live?
At the execution boundary responsible for that effect, using a stable trusted operation identity and concurrency protection.