# Release an unseen test

Use these steps only at the exact release time printed on a week page. Before Action 01, that block's **EXACT RECORD FILES** panel prints the complete test filename. It shows commands for your selected shell and the private save steps. Create the empty file there on purpose. Write the details named below. Leave only the test body blank until release time.

> **What this teaches:** I can take a fresh, timed sample of my own skill without getting a solution or hidden coaching.

For Weeks 1–2, start the session clock before these steps. File creation and prompt preparation use the published preparation slot.
The assessment has its own timer inside that session. Release the task only when its test slot begins.
Task release is not coaching. Close the release tool before answering. Keep all timed work free of AI help.

## Before opening the model

1. Create the empty private test file named by the week page.
2. In that file, write the week, task type, planned release time, and published pass line. Quizzes have scores, not a pass line.
3. Open a fresh Claude session. Attach no project files and no earlier attempts.
4. Fill in every bracket below from the published week page. Don't include source code, private data, or another answer. The release nonce is a one-time code that makes this prompt unique.
5. Paste the completed prompt once.
6. Save the returned task in the private test file. Write the real start time. Close Claude. Start the clock.
7. Don't reopen any AI tool until your answer or commit and your stop time are written down and final.

## Prompt to paste

You are releasing one unseen Engineered Bootcamp test. Produce only the test content. Do not inspect a repository, run tools, or write code. Do not give a solution, give hints, explain an approach, or score work that does not exist yet.

Task type: `[20-minute quiz / 45-minute change task / 60-minute weekly test]`

Week and route: `[fill in]`

Can-do targets being tested: `[fill in]`

Required Java concepts: `[fill in]`

Concepts not yet taught, so not allowed: `[fill in]`

Published rubric and pass line: `[paste unchanged]`

Already-seen examples to avoid: `[fill in]`

Made-up project setting: `[fill in]`

Release nonce: `[current local date/time plus six random characters you choose]`

Create a fresh task that fits the named time limit and works on any computer. It must need no purchase, no deployment, no outside account, and no missing dependency. Use only made-up data.

For a 20-minute quiz, give one small behavior change with inputs, expected outputs, edge cases, and a fixed `/10` rubric.

For a 45-minute change task, give one small feature or bug. Include a user story, three measurable acceptance criteria, one clear non-goal, and the published `/100` rubric.

For a 60-minute weekly test, give exactly eight short trace, predict, or diagnose items, plus two small write tasks. Use the published weights unchanged.

Output only these parts:

- the test ID;
- the start instructions and time limit;
- the task;
- supplied examples, if needed;
- the rubric and pass line;
- the proof to save.

End with: `No solution or hint has been provided. Close this session before starting.`

## After the clock

In the same private test file, write the stop time, the result, and where your proof is. Also write any unfinished work, honestly. Reopen the file. Before you ask for review, check that the task, start time, stop time, and result are all there.

## Explain it back

Explain what a fresh prompt, a fixed pass line, a closed AI window, and a final stop time can test. Then explain why normal guided work cannot test it.