- Published
- Updated
Avoiding copies of real participant data can support privacy during testing. Yet overly simple synthetic examples may miss important defects because they never exercise long comments, missing values or realistic relationships between records.
Preserve the meaning of data
Design test data around more than row counts. Create a coverage list for critical field boundaries, record relationships and expected exceptions. Synthetic records should be recognisable and prevented from entering genuine study workflows. Removing direct identifiers does not automatically establish adequate privacy in every context; responsible parties should assess the proposed data-use arrangements separately.
Worked example
For example, every test participant has one visit and a short name. A production participant with multiple visits later produces incorrect links during transfer. The team adds repeated visits, corrected results and delayed delivery to a synthetic participant and writes the expected relationships in advance. Targeted complexity is more useful here than additional random rows.
Review checks
- Identify the synthetic example representing each critical requirement.
- Cover long text, blanks, boundaries and linked records together.
- Prevent test identities being used as genuine production participants.
Acceptance and boundaries
Acceptance means the dataset represents the defined business risks and its expected results can be understood independently. A synthetic-data generator should not create incorrect expectations; review critical examples. Unnecessary real data do not strengthen test evidence. Keep the example version and the tests in which it was used retrievable when reconstruction is needed.
Frequently asked questions
- Do more test rows always mean better coverage?
- Volume may test a different question; missing critical scenarios can remain absent even in a large dataset.
- Are de-identified source data synthetic?
- No. Distinguish derived source data from wholly invented examples and keep their conditions of use explicit.