Data Quality Guardian — Catch Broken, Missing & Suspicious

Data quality brief — illustrative four-row sample

Purpose/grain: one current row per order_id under the supplied rule. Full source authority, update versions and reporting criticality are unknown.

Finding 1: O1 appears twice with identical values. Violates supplied current-order uniqueness rule, pending confirmation that rows are not distinct legitimate events. Preserve raw rows; do not delete automatically.

Finding 2: O2 amount -5 violates supplied non-negative rule. It could reflect different business semantics such as a return only if the contract allows that; do not change the supplied rule without owner review.

Finding 3: O3 blank amount is unresolved. No rule defines null as invalid or zero. Quarantine or flag for owner decision; never silently coerce it to zero. Currency values are present in all four rows, but allowed currency domain is not supplied.

Priority: verify whether these rows feed financial or other consequential reporting, determine blast radius and halt publication only where agreed policy requires. Four sample rows cannot establish production prevalence or root cause.

Prevention controls: documented grain and uniqueness scope; explicit amount/null semantics; approved currency domain; source-side validation where reliable; actionable rejection reasons; authoritative count/value reconciliation; owners and severity for each rule.

Investigation: trace duplicate generation and amount mapping through lineage, inspect representative authorized samples and compare control totals. Distinguish symptoms from proven cause.

Execution notes: no database accessed, rows corrected, alerts configured or report blocked. Findings apply only to the supplied fictional sample.