---
title: Prepare to start
description: Materials and decisions to prepare before configuration.
reviewedOn: 2026-09-08
sourceRefs: [evidence-lab-methodology]
---

## Result

You have a safe, bounded practice case with enough evidence to tell whether the environment worked. The case resembles real work but does not expose unnecessary personal, confidential, or credential data.

### Minimum case brief

Record the trigger, input files or links, required output format, known-good reference or reviewer, three to five acceptance criteria, permitted sources, prohibited data, and the person who can make the final domain decision.

## Actions

1. Select a recent, representative task whose correct or acceptable result you understand.
2. Copy only the permitted inputs into a dedicated folder; preserve originals outside the working area.
3. Remove or replace personal data, confidential reviews, credentials, unpublished material, and irrelevant metadata.
4. Write the required output format and three to five observable acceptance criteria before using an agent.
5. List the sources the environment may use, the sources it must not use, and whether web access is allowed.
6. Confirm the organisation's cloud-AI, retention, sharing, and account rules with the appropriate owner.
7. Record one likely failure that the practice case should reveal, such as a missing citation, wrong file version, or unsupported claim.

## Check

Give the folder and brief to a second person. They can identify the task, permitted inputs, expected artifact, acceptance criteria, and data boundary without opening an old chat or receiving additional access.

## If it fails

If no safe real example is available, create a de-identified structural copy that preserves the same file types, decisions, and acceptance criteria. If you cannot remove sensitive content reliably, stop and ask the data owner for a permitted case.

**Next:** [Choose a platform](/en/start/choose-platform/).
