Create a dataset

Datasets define the scope of data to be synthesized within an application. Each dataset is associated with a specific application and groups the tables required for a testing scenario.

Use this task to create a new dataset in Synthetic Data.

Prerequisites

• An application with at least one reference connector must already exist.

• Discovery must be complete for the application.

Steps

1. From the top navigation bar, click Synthetic, then click Datasets in the left navigation pane.

2. In the upper‑right corner of the Datasets page, click + Create Dataset.

   The Create Dataset wizard opens.

3. On the Select Application page, choose the application this dataset belongs to:

   • In the Search applications… field, search for or browse available applications.

   • Click the application card to select it.

     The available connectors for the selected application are displayed and used as data sources.

   • Click Next.

4. On the Describe Test Data page, choose how to identify tables for the dataset:

   • Describe your test data (recommended) - Describe your testing scenario in plain language. Synthetic Data uses AI-assisted discovery to identify relevant tables.

   • Select tables manually - Browse and select individual tables directly from the data source.

   • If you choose Describe your test data:

     • In the Describe your testing scenario field, enter a description of your testing scenario. For example, I need customers with addresses, linked to orders containing payment and line item data for end‑to‑end checkout testing.

     • Optionally, click one of the example prompts (Checkout flow, Customer data, or Product catalog) to populate the field.

     • Click Find Matching Tables.

       Synthetic Data uses AI-assisted discovery to analyze schema metadata and identify tables that are relevant to the entities and relationships described in your testing scenario. The wizard then advances to the Review Tables page with a suggested set of tables grouped by connector, each displayed with a match score.

       Only tables that meet the configured confidence threshold are included in the recommendations. The match-score column label and tooltip explain that the percentage indicates how closely each table matches your description, and the wizard header displays the confidence threshold used for the search. If an expected table is not listed, click Show All Tables on the Review Tables page and select it manually. For more information, see Configure Synthetic Data settings topic.

   • If you choose Select tables manually:

     • Click Next. The wizard advances to the Review Tables page with all tables from the data source listed, grouped by connector, without match scores.

5. On the Review Tables page, review and refine the tables included in the dataset. Tables are grouped under their connector (for example, DB connector 1 (9), where the number indicates the table count).

   • Select the connector‑level checkbox to include all tables in that group, or expand the group to choose individual tables.

   • Use the Search tables… field to filter the table list.

   • Select or clear individual table checkboxes to include or exclude tables. The selection counter (for example, 3/9 selected) updates as you choose.

   • Click Select All to include all listed tables, or Clear All to remove all selections.

   • If you described your test data, each identified table also displays a match score (for example, 86%), indicating how closely it matches your description. The page displays the confidence threshold used for the search, such as Showing tables with ≥60% confidence. Click Show All Tables to display every table in the data source, including those below the threshold.

   • To revise an AI description, expand the YOUR DESCRIPTION section and click Edit to change it and run the search again.

   • Review the relationship selections before continuing:

     • Related parent tables can be included automatically when a selected table references them.

     • Child tables might not be selected automatically. If the warning indicates that child tables are not selected, click Add child tables to include them. Omitting required child tables can cause a foreign key constraint violation when a job uses the replace data option.

     • The relationship indicator beside a table shows the related table path, such as ORDERS -> CUSTOMERS -> EMPLOYEES.

   • Review the Automatic Relationship Detection section. It explains that foreign key relationships between selected tables are detected automatically and maintained to preserve referential integrity. Tables without a match score might be included because selected tables depend on them.

   • Click Next.

6. On the Dataset Info page, provide the following details:

• Name: Enter a unique name for the dataset.

• Description (optional): Enter a free‑text description to help identify the dataset.

• Tags (optional): Add one or more key‑value tags.

  • Enter a value in the Enter Tag Key field.

  • Enter the corresponding value in the Enter Tag Value field.

  • Click + Tag to add the tag.

  • Repeat to add additional tags.

• Click Next.

7. On the Review & Create page, review your dataset details, including the dataset name, description, tags, associated application, and selected tables.

   • To make changes, click Back.

   • To create the dataset, click Submit

The dataset is created and appears on the Datasets page. You can open the dataset to configure tables, relationships, generators, and data volume, and to create jobs for synthetic data generation.

Generator assignment during dataset creation

When you create a dataset, Synthetic Data automatically assigns generators to each field using a three-layer discovery cascade. This process identifies sensitive columns and assigns semantic generators that produce realistic data. For detailed information about how generators are assigned and what the different assignment methods mean, see Synthetic Data discovery methods.

If the application uses Light Data Sampling, Data-Based Discovery analyzes sampled source values during the dataset pull. After the pull completes, review the assignments in the dataset Fields view and override any assignment that does not represent your test data.

After creation, you can review and modify generator assignments by opening the dataset and navigating to the Tables tab, then viewing the Fields for each table.