What does a 'data contract' actually look like in practice?

Document the grain decision, most BI bugs turn out to be grain bugs.

Open source tools like Great Expectations or Soda can enforce the contract at runtime, failing the pipeline if incoming data violates the agreed shape instead of only catching it at review time.

Simple way to think about it: caching is a bet that you'll ask the same question again soon. If you don't, you're just paying rent on memory for nothing.

In practice it's a schema (Avro, Protobuf, or JSON Schema) checked into the producer's repo, versioned, with CI that fails the producer's build if a breaking change ships without a version bump.

Any downside to this approach with incremental models?