SynData
Synthetic Data Validation Pilot
Named deliverable Synthetic Data Validation Scorecard
The pilot starts with one dataset, one intended use, and a written threat model. I choose fidelity, utility, and privacy checks that fit those facts instead of pushing every dataset through the same score. The work runs against the real and synthetic data in an agreed environment. You get the code, results, limitations, and a recommendation to use the data as planned, revise the generator, restrict the use, or stop.
What you get
- A written use case and threat model
- Fidelity checks suited to the data type
- Utility tests tied to the downstream analysis or model
- Privacy attacks chosen for the realistic disclosure risks
- Reproducible scripts, configuration, and seeds
- A scorecard that keeps the evidence separate instead of hiding it inside one number
- A use, revise, restrict, or stop decision memo
- A list of tests to repeat when the generator, source data, or intended use changes
Who it's for
Teams deciding whether an existing synthetic dataset or generator is fit for a defined use