Data Lineage Vocabulary: How to Talk About Data Provenance and Impact Analysis

When something breaks in a data pipeline, the first question is always “where did this data come from?” and “what else does it affect?” Answering those questions confidently — in English, in a cross-functional meeting — requires a specific vocabulary that most data engineers pick up slowly through trial and error. This guide accelerates that process.

Understanding Lineage Concepts

Data lineage is the ability to trace data from its origin through every transformation and movement to its final destination. It answers the question: “How did this number end up in this dashboard?”

Upstream lineage refers to everything that feeds into a given dataset — the sources, transformations, and jobs that produced it. Downstream lineage is everything that depends on it — reports, models, and other tables that consume it. Engineers say: “Before we change this column type, let’s check the downstream lineage — there are at least 14 dependent tables.”

A lineage graph is the visual representation of these relationships. It shows nodes (datasets, tables, jobs) and edges (the data flows between them). You will hear: “Pull up the lineage graph for the orders table — I want to see the full dependency chain before we run the migration.”

Lineage can be captured at different granularities. Table-level lineage shows which tables feed which other tables. Column-level lineage (also called field-level lineage) is far more precise — it tracks how individual columns are derived, transformed, and mapped across systems.

Data Provenance and Governance

Data provenance is the broader concept of recording the complete history of a piece of data — where it was created, who modified it, when, and how. While lineage focuses on flow, provenance focuses on accountability and auditability.

A data catalog is the system that stores and surfaces metadata about your data assets — schema, sample data, ownership, and lineage. Tools like DataHub, Atlan, OpenMetadata, and Alation are common in the industry. A data dictionary defines the meaning of each field in a dataset. A business glossary maps business terms (like “active customer”) to their technical definitions.

Data stewards and data owners are the people responsible for data assets. The owner is accountable for the data’s quality and usage; the steward is the day-to-day custodian who maintains definitions and monitors quality. In meetings: “Tag the data owner for this dataset — they need to approve the schema change before we propagate it downstream.”

Impact Analysis and Breaking Changes

Impact analysis is the process of understanding what will break — or change — if you modify a dataset. This is where lineage becomes operationally critical. Before a schema change, engineers run impact analysis: “The impact analysis shows 6 downstream consumers of this column — we need to coordinate the migration with all of them.”

A breaking change is any modification to a dataset’s schema or semantics that will cause downstream consumers to fail or produce incorrect results. Renaming a column, changing a data type, or removing a field are common examples. Tools like Monte Carlo and Great Expectations can detect these automatically.

In GDPR contexts, data lineage is essential for implementing the right to erasure (right to be forgotten). When a user requests deletion, you need to trace every system that holds or derived data from their records. Without lineage, erasure is guesswork.

Common Phrases in Lineage Discussions

• “We have no column-level lineage in this part of the warehouse — it’s all undocumented SQL transforms.”
• “The catalog shows this field as deprecated, but three active models still reference it.”
• “We need to propagate the lineage metadata from the dbt models into DataHub automatically.”
• “Can you add a business glossary entry for ‘conversion’? Marketing and engineering use different definitions.”

Practice

Find one table or dataset in your current project that has no documented lineage. Map its upstream sources and downstream consumers manually, then write three sentences describing its lineage in English using the vocabulary from this article. This is the foundation of communicating data architecture decisions clearly.