---
title: Repeatable data and behavior
---

# Repeatable data and behavior

Use real Stripe and OpenAI SDKs with exact released twins. Node 22.6 or newer and npm are required; no platform account, vendor credential or source checkout is needed.

Its public example files are generated from the canonical [shape a test tutorial](../../docs/guides/shape-the-world-for-a-test.md), so the runnable code has one source. Files: [package manifest](../../docs/guides/shape-the-world-for-a-test.md#separate-records-from-behavior), [environment names](../../docs/guides/shape-the-world-for-a-test.md#separate-records-from-behavior), [workflow](../../docs/guides/shape-the-world-for-a-test.md#separate-records-from-behavior) and [handler](../../docs/guides/shape-the-world-for-a-test.md#separate-records-from-behavior) and [scenario wiring](../../docs/guides/shape-the-world-for-a-test.md#wire-the-handler-for-this-released-runtime). Install, init, review the selected vendors, then save the handler after init as that guide explains.

The workflow asserts a stored customer and frozen timestamp, a one-time 429 refusal, a scripted retry answer and customer read-back. Reset and restore the clock before the second walk. The guide owns the exact commands and expected output.

The model answer is authored behavior; no model quality is measured. The score is the named app assertions, not full vendor surface coverage. Restart rearms the once-handler; `up` alone resumes stored data. Use a task-owned disposable branch for reset.