How should a reseller check grounding and disclosure risk before licensing generated data onward?
Practical steps a reseller can take to inspect a generated sample for grounding and disclosure risk, and what not to assume from the fact that data is synthetic.
**Does generated data remove disclosure risk?**
No. Generated data can still resemble real people, organisations or documents by chance or through the way it was produced. Treat anonymity and privacy compliance as things to inspect, not assume. MirrorFirm output is generated, and it is not real business records.
**What should I check?**
- Search names, addresses, identifiers and phone numbers against public sources for accidental matches. - Read free-text fields for distinctive details that point to a specific organisation. - Check identifier styles, such as `APP-2187`, are clearly non-real formats. - Look for near-duplicate records that suggest memorisation of a single source. - Ask how the generator was informed by direct observation of how businesses operate, and what was done to keep that separate from output.
**What should I ask MirrorFirm?**
Ask for a description of the grounding approach, the known limits of the sample, and what documentation you may pass on.
**What should I tell my customers?**
State that the data is generated, describe the checks you ran, and recommend they run their own. Do not describe the data as anonymised, compliant or approved by any regulator.
Synthetic does not mean risk-free, so resellers should run their own inspection and describe results honestly in listings and onward terms.
- Audience:
- Licensed resellers
- Content type:
- FAQ
- Buyer question:
- How to evaluate it
Content release: r-a220eb6e099b243b
Related questions
- What trace format and grader hooks make conditional-approval evaluation automatic?
- How do I check that generated tasks are hard enough and do not leak the answer?
- How should I score an agent on conditional approvals beyond a single pass rate?
- Which checks should I run on a sample before committing to wider use?
- What does a failed evaluation run look like when an agent binds a policy too early?
- How do I wire graders and reset into a conditional-approval environment?