Self-assessment

Select the statement that best describes the team today, one per chapter. The result shows a level for each chapter and the chapters to read first. Answers are stored in this browser only.

1. Purpose Is the purpose of synthetic files stated?

Not yetNo synthetic files, or files produced without a stated purpose.FoundationalA purpose statement exists and one file is produced for code development and teaching.AI-readyFiles for the main surveys are produced from demand with intended uses recorded and labelled everywhere.AI-nativeFiles are produced at each release as a pipeline stage with their use tracked.

2. Method How is the synthesis method chosen?

Not yetWhatever tool was at hand.FoundationalA sequential method with the seed recorded.AI-readyMethods are chosen by a written rule from data and uses, compared on the same measures, with parameters and alternatives recorded.AI-nativeRelational generators for multi-table surveys, tuned by the target analyses, re-run at each release.

3. Utility How is utility measured?

Not yetIt is not.FoundationalMarginals and two example analyses compared.AI-readyMarginals, associations, and the target analyses of the intended uses, against thresholds, attached to the record.AI-nativeThe report runs as a pipeline stage and is published with the file.

4. Risk How is disclosure risk measured?

Not yetSynthetic is assumed safe.FoundationalExact copies are checked and a file with copies goes back to the method.AI-readyClosest-record distance and attribute inference are compared with baselines, limits are set by the disclosure control unit, and the review is recorded.AI-nativeMembership inference and formal methods are added where the tier requires, as a pipeline stage.

5. Relational How are linked tables handled?

Not yetTables are synthesized separately or not at all.FoundationalThe person table is synthesized with household variables; integrity and structure are checked.AI-readyA relational generator is used; survey rules are added to the check; structural distributions are compared.AI-nativePanels are synthesized wave by wave with transitions checked; checks run as a pipeline stage.

6. Release How is a release documented?

Not yetFiles are shared without a record.FoundationalA synthesis record from the template; the file is labelled and linked to the real survey.AI-readyThe record is checked; reports attached; the release decision recorded with reviewers, dates, tier, and review date.AI-nativeRecords are generated from the pipeline and published with the reports.

7. Workflow How do synthetic files enter the workflow?

Not yetThey sit in a folder.FoundationalOffered on the survey page with guidance; downloads counted.AI-readyOffered in the application and enclave workflows; effect measured; feedback collected.AI-nativeServed through the same interfaces as public data, labelled, with use tracked.

0 of 7 answered

Interpretation of the result

• Chapter 1 first: a purpose statement and a label take a day and prevent the misuse that makes synthetic data a risk.
• Chapters 3 and 4 before any release: a file nobody measured is a file nobody can vouch for.
• Chapter 7 is where the value appears; a measured file nobody is offered changes nothing.