Data Engineering & Analytics

A data quality score may improve while users report worse results if the score measures completeness or format rather than correctness and usefulness. The checks may also exclude a changing source or focus on records that are easy to validate. What to check Compare score definitions and coverage with the affected user workflows. Review validation...

A source backfill can correct missing history but also change previously published reports. The effect depends on whether downstream jobs reprocess old periods, deduplicate records, or preserve snapshots. What to check Compare the backfill range and source identifiers with transformation watermarks, late-arriving data rules, and report refresh dates....

A data model update can alter joins, default filters, calculated fields, or relationship direction. Reports may still render normally while counts shift or some records disappear. What to check Compare the old and new model definitions, changed relationships, calculation logic, filter defaults, and report refresh time. Trace a few known records from...

Reports may change after customer statuses are reclassified because filters include active, closed, or historical accounts differently. The underlying transactions may be unchanged while the population counted by the report shifts. What to check Compare status history, report filters, effective dates, joins, and whether metrics use current or...

A source system can add, rename, or change a field while continuing to send data. Pipelines may ingest the new shape but silently lose values, miscast records, or skip quality rules that were written for the previous schema. What to check Compare schema versions, null and type distributions, rejected rows, transformation warnings, and downstream field...

A currency conversion update can change historical totals when reports recalculate past transactions using a new rate or a different rate date. The change may be correct, but it needs to be distinguishable from new business activity. What to check Check the conversion source, effective date, rounding rule, base currency, report refresh, and whether...

A daily report can miss events from the prior evening when its time window uses UTC, a local business day, or an ingestion cutoff that differs from the source system. The records may appear the next day and shift totals unexpectedly. What to check Compare event timestamps, ingestion timestamps, time zones, daylight-saving behavior, and the report’s...

Deduplication can remove legitimate records if the matching key is too broad, or fail to remove duplicates if identifiers change. A lower row count is not automatically a better dataset. What to check Compare deduplication keys, ordering rules, null behavior, source identifiers, and before-and-after samples. Identify whether distinct events share a...