Datasets define the scope and configuration of synthetic data generation for an application. A dataset is always associated with one application and is built using metadata discovered from the application’s reference connectors.
With a dataset, you can:
-
Select the tables required for a testing scenario.
-
Use AI-assisted discovery to identify relevant tables based on a natural-language description of your testing requirements.
-
Review and adjust generator assignments for fields.
-
Control the amount of data generated.
-
Manage relationships between tables to preserve referential integrity.
AI-assisted discovery analyzes schema metadata and recommends tables that are relevant to the described testing scenario. The recommendations are based on configurable confidence thresholds and can be refined by manually reviewing and selecting tables.
When creating a dataset, you can either select tables manually or describe your test data needs and let the system suggest relevant tables.
Once created, each dataset automatically has generators assigned to all fields through a three-layer discovery process that identifies sensitive data and assigns semantic generators. You can review, compare, and modify these assignments as needed. For details about how generators are automatically assigned to fields, see Synthetic Data discovery methods.
For correlated values, you can group multiple fields from the same table under one multi-output generator. The generator runs once per row for the group so related fields are generated together. Cross-table correlated generation is not supported. For configuration steps, see View a dataset.
Once configured, a dataset is used to create jobs that generate synthetic data into one or more target destinations.
The Datasets grid lists the datasets for an application and provides access to each dataset's jobs, tables, fields, and relationships.
Long-running dataset operations support cancellation. You can cancel a dataset pull, duplication, or structure deletion from the available action. API users can call POST /synthetic/datasets/{id}/cancel-pull, POST /synthetic/datasets/{id}/cancel-duplicate, or POST /synthetic/datasets/{id}/cancel-delete-structures. Canceling a dataset pull reverts newly created Structure rows without deleting the dataset. Canceling structure deletion rolls back the deletion transaction. If duplication is canceled after the target dataset is created, DCT removes the target dataset. If the service restarts while one of these operations is in progress, the operation and its associated job are marked Error with a restart message.