You're designing a new pipeline to capture events from human feedback interfaces and make them available for downstream analysis and training workflows. You want a design that can support real-time ingestion, validation, and replay without creating duplicate records or breaking when payloads change.
How would you design a data pipeline to handle real-time ingestion from human feedback interfaces?