People mix these up. In data engineering interviews, a clean split helps:
Validation asks: "Does this data meet the rules we defined?" You check schema, nullability, ranges, enums, referential integrity, and freshness thresholds against expectations.
Verification asks: "Is this data correct relative to a trusted source or business truth?" You reconcile row counts, sums, and samples against upstream systems, invoices, or known fixtures.
validation: order_id NOT NULL, amount > 0, status in allowlist verification: sum(fct_orders.amount) for 2026-09-05 == Stripe day total
Tiny story
A pipeline loads 10,000 orders. All not_null and unique tests pass (validated), but revenue is $50k short vs Stripe (failed verification). Schema rules alone were not enough.
How teams use both
- Validation runs cheaply on every build (dbt tests, Great Expectations).
- Verification runs as reconciliations, audits, or finance close checks.
Interview tip: "Validation = rules; verification = truth against a reference." Give one example of each.