Data Quality Lab: Make Errors Visible
Test checks on synthetic records and separate auto-fixable issues from cases requiring review.
Status: interactive lab using fictional synthetic data and documented rules.
Distinctions the lab must preserve
Intentionally messy, entirely synthetic supplier and item data.
An explicit validation criterion with a testable expected outcome.
Ambiguous records stay visible and unresolved.
The before-and-after lab is interactive and uses only synthetic data.
Interactive rule checks are enabled.
Explore messy synthetic records
The sample rows deliberately contain gaps and inconsistencies: missing identifiers, duplicates, key collisions, unusual dates, inconsistent units and currencies. Every original value must remain traceable after any proposed transformation.
INTERACTIVE LAB
15 fictional records and 8 independently selectable rules. No user uploads.
Synthetic sample data – loading the module.
Select and run validation rules
Visitors can select individual validation rules and inspect their outcomes. Checking does not necessarily mean correcting. Every rule needs a precise definition, execution order, and positive and negative test cases.
Missing and duplicate keys
Detect incompleteness, reuse and key collisions as distinct issues.
Units and currency
Only normalize with known mappings and documented conversions.
Dates and validity
Expose invalid or conflicting effective-date ranges.
Compare before and after
The comparison shows the original value, proposed transformation and resulting value. Consistent formatting may be automated under a fixed rule; an ambiguous product or supplier identity must not be silently overwritten.
Keep the original visible
Source row, field and unique synthetic example identifier.
Explain the outcome
Valid, flagged, ambiguous or requiring human review.
Only approved changes
Separate deterministic normalization from unapproved suggestions.
Review unresolved cases
An effective quality log retains unresolved entries, priority, triggering rule and pending decision. A review queue is an appropriate output, even if it makes the dashboard look less complete. Business approval stays with an accountable reviewer.
Method and limitations
This lab demonstrates traceable quality rules using synthetic records only. It is not a production data-cleaning service, certification or legal audit. Ambiguous cases remain in the human-review queue.
Which issues can be auto-corrected?
Only deterministic, pre-defined normalizations should qualify. Ambiguous business cases require review.
What happens to ambiguous rows?
They must stay visibly unresolved, with the rule and reason recorded for a reviewer.
Clarify the data and scope together
A scope discussion starts with specific sources, checks and boundaries, not unverified outcomes.