What Are Data Contracts and SLAs?

The Core Problem: In large engineering organizations, teams ship features constantly. An application team might rename user_id to customer_id or change amount from cents to dollars. Without explicit agreements, downstream data teams discover these changes when dashboards break, Machine Learning (ML) models silently degrade, or pipelines fail at 2 AM. Data Contracts Define Structure: A contract for a user_signup event stream might specify that every event must contain a non null user_id, a created_at timestamp in Coordinated Universal Time (UTC), and a country_code following ISO 3166 standard. It also defines evolution rules: fields must remain backward compatible for at least 90 days. SLAs Define Performance: SLAs translate to Service Level Objectives (SLOs) with measurable Service Level Indicators (SLIs). For example, "95 percent of events arrive in the data warehouse within 5 minutes, 99 percent within 15 minutes" or "daily orders table is available by 03:00 UTC with 99.9 percent success rate over a quarter." The Analogy: Think of data contracts like APIs for data. Just as a REST API defines endpoints, request formats, response schemas, and uptime guarantees, data contracts define what data is produced, in what format, with what quality, and how reliably. This transforms data pipelines from fragile, undocumented systems into production grade infrastructure with explicit expectations.