Data contracts: the fix for pipelines that break on Mondays

Every data team has a Monday morning where a dashboard is wrong and nobody knows since when.

Make the schema a public interface A data contract is a versioned schema plus semantics, ownership, freshness expectations and a deprecation policy. It lives in the producer repository and is enforced in the producer CI.

Fail at the source Breaking changes should fail the producing service build. If they only fail downstream, you have moved the cost onto the least-informed team.

What we include - Field types, nullability and allowed value ranges - Owner and on-call route - Freshness and volume expectations with alerts - A documented deprecation window before removal

Pay for the observability Row count anomalies, null-rate spikes and distribution drift catch the failures schemas cannot. Contracts stop structural breakage; monitoring stops semantic breakage.