# First-use exercise

[中文](task.zh-CN.md) · [Feedback form](feedback.en.md) · [Version notes](versions.en.md)

Try these five steps as a technical user who has not seen the project before, using only the packaged documentation. Allow roughly 30–45 minutes; stop and record a blocker if needed. Use synthetic data and built-in scripts only. No model, account, paid service, or customer material is required.

1. **Start.** Record the archive version, SHA256, operating system and Python version. Follow the [quick start](../quickstart.en.md) to create a local environment. Record extra steps and messages encountered before the first successful run.
2. **Understand the rules.** Explain in your own words when each task should stop or execute, and how consent, contact frequency, refunds and budget affect the operation. Mark wording that requires a guess. Budget is a ceiling: each coupon is 1000 cents and fewer eligible members need not consume the full 3000-cent budget.
3. **Run both tasks.** Run both reference tasks on the same data instance. Open the reports and identify their different final states. With 0.5.0, select one additional data instance, repeat both tasks and confirm the recorded instance. With 0.4.1, skip this additional-instance step and keep the old archive unchanged.
4. **Explain one failure.** Run the intentional refusal example in the quick start. Explain the failure in one or two sentences and cite evidence paths in the matching JSON. Distinguish an incomplete task from an actual violation; a score alone is not an explanation.
5. **Assess integration obstacles.** Read the [integration overview](integration.en.md) and task/tool specifications and list missing information for connecting your own agent: receiving tasks, issuing tool calls, preserving state, stopping, and reading reports. This step is reading and discussion only; do not start a model, send requests or provide credentials.

There is no need to demonstrate that the tool is useful. Record where you got stuck, what you expected, what happened, whether someone helped, and whether these errors would matter in your work. An anonymous participant code is sufficient. Do not attach customer data, access tokens or personal contact details.

The inviter arranges participation and collects feedback. Dates require agreement between participants; this page neither books a session nor submits any response automatically.