Synthetic Data

Synthetic data is simulated or generated records that were not drawn from real filed matters. Production models that support claims about courts and regulators are not trained on it.

Synthetic records can be useful for testing software. They cannot support a claim about how real courts, real judges and real regulators behave, because the model has learned from invented behavior. If the data does not exist, the model is not built.

The rule extends to imputed records. Filling missing fields with generated values changes the population the model learns from and hides the very sparsity a sufficiency floor is meant to expose.

The exclusion applies to training populations. It does not prevent synthetic inputs from being used in clearly labeled stress tests or illustrations, provided they are never presented as evidence about real outcomes.