2026-11-07 · 4 min read
How to Define Test Data Boundaries for a Laboratory AI Pilot
Decide what test records can enter a laboratory AI pilot, what must be excluded and how to check that the boundaries hold.
Start with the question the test must answer
A test of whether a summary preserves units does not need a copy of production records. Write down the behaviour to test, then build a synthetic record with the fields that matter. Put a clear fictional marker in each record so staff cannot confuse it with real work.
List allowed inputs and common surprises
Specify approved fields, file types and environments. Look beyond the obvious fields: a spreadsheet may contain hidden tabs, comments or filenames that reveal information. A synthetic example should be written from scratch, not made by renaming a patient or sample identifier in a real record.
Give reviewers a small boundary challenge
Use an approved fictional file with an extra column, a blank unit field and a duplicate reference. Decide beforehand which records should be rejected or flagged. Check that the test captures what happened and who reviewed it without copying test material into an unapproved tool.
Record the decision and the limits
Keep the dataset version, scenario, expected behaviour, observed behaviour and review outcome. If approved access to representative data becomes necessary, treat that as a separate decision for qualified laboratory, quality and information-governance staff. Do not silently expand the test because synthetic results look promising.
Does synthetic testing validate clinical use?
No. It can expose workflow problems and check safeguards under limited conditions. Qualified staff must determine validation and approval requirements. This guide addresses laboratory operations, not diagnosis, treatment or clinical decisions.