Skip to main content
Start with small synthetic fixtures whose expected results you can calculate independently. This guide runs locally; it does not require a hosted Synthetic Data Studio. The earlier Studio host and API examples were not backed by a published service contract and have been removed.

Generate a repeatable fixture

Use Node.js 20.20.0+. Save this as generate-fixture.mjs in a new directory:
Success: the script writes orders.fixture.json and prints the expected count and total. It refuses to overwrite an existing file. Prices are represented as decimal strings; the calculation uses integer cents rather than floating-point addition. These are input fixtures, not API request bodies or persisted commerce records. Map them to the exact binding’s schema before importing them. The orders quickstart shows the tested customer, inventory, and order creation sequence.

Cover decisions, not just volume

Use the error-handling tests for concrete rejection checks and agent evaluations for expected conversational behavior.

Keep a test record

Record the fixture version, environment, package version, expected results, and the IDs created by the run. Use separate databases and workspaces for tests. Synthetic data can still trigger real delivery or payment actions if connected to live tools; configure the execution boundary.
Synthetic fixtures are not evidence of production representativeness or a privacy guarantee. Review any imported source data and independently verify what connected tools can execute.

Continue

Run the first local operation, then add the edge cases above to the integration test plan. Keep fixtures and their expected results under version control so failures can be reproduced.
Last modified on September 21, 2026