Data Lineage Engineer — Interview Questions

Option B is strongest: it uses a concrete example (executive dashboard revenue figure) rather than an abstract definition, provides the specific stakeholder-friendly description of what column-level lineage reveals (source field, filter step, aggregation, currency conversion), contrasts the with-lineage and without-lineage investigation experience, quantifies the business value (days to hours), and gives the exact sentence pattern for communicating a lineage investigation result. The 'I traced the data from...' pattern is the key deliverable sentence. Lineage vocabulary: Column-level lineage — tracking the origin and transformation history of a specific column or field value. Data tracing — following a value backward through its transformations to its source. Transformation step — any SQL, Python, or dbt operation that modifies data as it moves through a pipeline. Root cause analysis — the investigation process for identifying why data is incorrect or unexpected. Options C and D are accurate but lack the concrete example and the contrast between with-lineage and without-lineage investigation experience.

Option B is strongest: it provides four named steps, introduces the critical insight that a rename three hops downstream can break a seemingly unrelated report, introduces the breaking/advisory/no-impact severity classification with concrete examples of each (including the dangerous advisory case of a currency rename), and provides the exact impact report sentence that turns lineage data into a decision. The 'three hops downstream' insight is what separates a lineage expert from someone who only checks direct consumers. Impact analysis vocabulary: Impact analysis — the process of identifying all downstream assets affected by a change to an upstream data asset. Dependency tree — a hierarchical representation of direct and indirect downstream consumers. Breaking change — a modification that causes downstream logic to fail or return incorrect results. Advisory impact — a modification where downstream logic continues to run but the semantic meaning changes in a way that may not be immediately visible. Migration window — the agreed time period during which downstream consumers migrate to the new schema. Options C and D are accurate but lack the three-hop insight and the severity classification examples.

Option B is strongest: it grounds the tracing process in a specific realistic scenario (churn rate discrepancy), uses named lineage vocabulary (starting node, lineage traversal, divergence point), introduces the row count comparison technique at each transformation step, identifies a logic change (not a data issue) as the root cause — which is a realistic and non-obvious finding — and provides the exact communication sentence that separates cause (logic change), from data (source is clean), and specifies the fix. The 'source data is clean' distinction is critical for communication — it prevents unnecessary source system investigation. Data quality vocabulary: Lineage traversal — the process of following a lineage graph upstream or downstream from a starting node. Divergence point — the transformation step where a metric's value first deviates from expected. Root cause — the original source of a data quality issue. Logic change — a modification to transformation code that changes the computed output without changing the source data. Reprocessing — re-running a pipeline over historical data to apply a corrected transformation. Options C and D are accurate but lack the specific realistic scenario and the 'source data is clean' distinction.

Option B is strongest: it explains why each component is necessary (not just what it is), introduces the 'impact first' principle and justifies it (decision-making context is most urgent), provides concrete example language for each section, introduces the critical 'affected vs. reliable data' component — which prevents collateral distrust of correct data — and explains why the monitoring commitment is more trust-restoring than the fix itself. That last insight is the mark of an engineer who has communicated data quality issues to business teams before and knows what actually restores trust. Data quality communication vocabulary: Impact statement — a description of the business-visible effect of a data quality issue, including quantification. Affected scope — the specific time period, channel, or dataset that contains incorrect data. Collateral distrust — the risk that a business team stops trusting correct data because they cannot distinguish it from affected data. Reprocessing — re-running a pipeline to correct historical data with a fixed transformation. Data quality alert — an automated notification triggered when data metrics fall outside expected ranges. Options C and D are accurate but lack the justification for each component and the monitoring commitment insight.

Option B is strongest: it opens with a frame the executive already cares about (accountability for numbers at board level), positions data governance and data lineage in their correct relationship (governance defines accountability, lineage makes it enforceable), explains three governance capabilities with the specific business scenarios where each matters (board presentations, regulatory audits, change control approvals), and provides the headline sentence that executives can repeat: 'governance as a policy versus governance as a technical control.' That distinction — policy vs. technical control — is the communication insight that separates a senior lineage engineer from a technical implementer. Data governance vocabulary: Data governance — the framework of policies, roles, and processes that define accountability for data quality, access, and use. Auditability — the ability to produce a documented trail of data origin and transformations for review. Data ownership — the assignment of accountability for a data asset's quality and maintenance to a specific team or person. Change control — the process of reviewing and approving changes to systems that may affect dependent data consumers. Technical control — a governance mechanism enforced by the infrastructure itself, rather than by human process adherence. Options C and D are accurate but lack the executive accountability framing and the policy-vs-technical-control distinction.