A take-home assessment should answer a specific hiring question. If the team cannot explain what it needs to learn, adding a larger exercise is unlikely to improve the decision. Begin with the evidence missing from the rest of the process, then choose the smallest task that can reveal it.
Choose one meaningful slice of the job
For a backend role, that might be extending a small service and explaining error handling. For an ML role, it might be reviewing an evaluation approach. For a platform role, it could be proposing a safe change to a simplified deployment process.
Avoid asking candidates to design an entire product, deploy a complex environment and write extensive documentation unless that effort is justified and transparently agreed. Use synthetic or public data, and keep the exercise separate from unpaid work your business needs delivered.
Write the rubric before the task
Decide which dimensions matter: correctness, reasoning, maintainability, testing or communication. Describe what acceptable evidence looks like for the role. If visual polish is not part of the job, it should not quietly determine the score.
Run the exercise internally. Check whether the setup works, the instructions are clear and the expected scope fits the advertised effort. Your own familiarity will make the task easier, so leave room for candidates to interpret an unfamiliar problem.
State the boundaries explicitly
Give a time expectation and say what to do when the candidate reaches it. Explain whether incomplete work with clear reasoning is acceptable. Specify which libraries, documentation and AI tools can be used, and apply the same policy consistently.
If tools are allowed, the follow-up can investigate understanding and ownership. Ask the candidate to explain a choice, identify a limitation or adapt a small part of the solution. Avoid changing the rules after the submission arrives.
Offer a fair way to demonstrate the skill
People differ in available time, access needs and comfort with assessment formats. Where practical, offer an equivalent live or discussion-based route that tests the same capabilities. Make it clear who candidates can contact if they need an adjustment.
For substantial work, consider compensation and obtain appropriate advice on the arrangement. More importantly, ask whether the assessment can be made smaller without losing useful evidence.
Use the submission as a starting conversation
Review the candidate’s decisions, not just the output. Ask what they would improve with more time and which assumption most affected the solution. Record feedback promptly against the shared rubric, then decide whether any unresolved gap actually matters to the role.