Event-Driven Architecture Planner

You are a distributed systems architect specializing in event-driven architecture. Help me design an event-driven system.

System description: [SYSTEM_DESCRIPTION]
Current architecture: [CURRENT_ARCHITECTURE]
Why event-driven: [MOTIVATION]
Scale requirements: [SCALE]
Consistency requirements: [CONSISTENCY_NEEDS]
Team experience with EDA: [TEAM_EXPERIENCE]

Design the event-driven architecture:

**1. Event Identification & Modeling**
- Domain events (what happened — past tense, business language)
- Event schema design (envelope + payload)
- Event versioning strategy
- Event catalog / registry

**2. Architecture Pattern Selection**
- Event Notification vs. Event-Carried State Transfer vs. Event Sourcing
- Recommendation for YOUR use case with rationale
- CQRS: Should read and write models be separated? Why/why not?

**3. Messaging Infrastructure**
- Broker selection (Kafka vs. RabbitMQ vs. SQS/SNS vs. Pulsar vs. NATS)
- Trade-off analysis for your requirements:
  | Criteria | Kafka | RabbitMQ | SQS/SNS |
  | Ordering | ... | ... | ... |
  | Throughput | ... | ... | ... |
  | Delivery | ... | ... | ... |
- Topic/queue design and partitioning strategy
- Consumer group design

**4. Data Consistency**
- Eventual consistency patterns
- Saga pattern for distributed transactions:
  - Choreography vs. orchestration recommendation
  - Compensation/rollback design
- Outbox pattern for reliable event publishing
- Idempotency strategy (deduplication keys)

**5. Error Handling & Resilience**
- Dead letter queues
- Retry policies (exponential backoff with jitter)
- Poison message handling
- Circuit breakers between services

**6. Observability for EDA**
- Distributed tracing across events (correlation IDs)
- Event flow visualization
- Consumer lag monitoring
- Alerting on processing failures

**7. Testing Strategy**
- How to test event-driven flows (contract testing, consumer-driven contracts)
- Local development setup
- Integration testing approach

**8. Migration Plan** (if moving from synchronous):
- Strangler fig approach for gradual migration
- Dual-write period handling
- Rollback strategy

**9. Anti-patterns to Avoid**:
- Event soup, chatty events, event-driven spaghetti, missing schemas