Data Source

Reis and Housley formalise this position in their data engineering lifecycle, where source systems occupy the first stage — generation — before ingestion, transformation, or serving. Their framework treats source characteristics (reliability, schema stability, data volume, latency) as upstream constraints🔒ConstraintStrategyA limit, requirement, or ceiling the product must respect, whether a self-imposed principle or an externally imposed boundaryView reference → that propagate through every downstream stage; a pipeline cannot be more reliable than the source it reads from.

How it works in practice

A subscription🔂SubscriptionSales & RevenueA recurring subscriptionView reference → business has three data sources feeding its warehouse: a Postgres billing database, a Stripe API, and a clickstream event topic. Analysts treat the billing database as the source of truth for revenue, because it is the system that records the actual charge. The clickstream is a source for engagement but never for money. When a finance figure and a product figure disagree, the first question is which source each was derived from, and the origin of the data establishes the starting point for tracing everything that follows.

Data source vs. its neighbours

• Data product🎁Data ProductData & AnalyticsA curated, reusable data assetView reference → is a curated, owned output built for consumption. A data source is raw input, the system a pipeline reads. A product is often assembled from several sources.
• Data pipeline🪣Data PipelineData & AnalyticsAn automated pipeline for data transformationView reference → is the process that reads from a source and transforms it. The source is where the pipeline starts; the pipeline is the work done after.
• Source of truth vs derived is a distinction within sources themselves. A system of record holds authoritative data; a derived dataset is computed from it and should never be confused for the original when reconciling numbers.

Reis and Housley draw this line explicitly: source systems are operational systems the data team does not own — transactional databases, SaaS platforms, IoT devices — and their data enters the pipeline but should never be confused with derived artefacts computed from it.

In the graph

In the Unified Product Graph, a data source is a hub in the data and analytics region, which fits because so many other concepts attach to it. A classification applies to it (data_classification_applies_to_data_sourceData Classificationapplies toData Sourcecross-domain), a contract governs it (data_contract_governs_data_sourceData ContractgovernsData Sourcecross-domain), a product ingests from it (product_ingests_from_data_sourceProductingests fromData Sourcehierarchy), and a source defines the metrics🔢MetricStrategyA unified metric that measures progress, health, or behaviour across the productView reference → drawn from it (data_source_defines_metricData SourcedefinesMetrichierarchy). That density is the point: pivot on a single source and you can see its sensitivity, its governing promises, and every metric that would move if it changed.