The take home test question
Candidates resent them, panels rely on them, and most of them are badly designed. When is a work sample worth the friction?

There are two honest positions on take home exercises. One is that they are the best predictor of job performance you can get in a hiring process. The other is that they push the cost of your hiring onto candidates who are already interviewing at four companies.
Both are true, which is why the argument never resolves.
The test that earns its place
A work sample is worth it when the job has a core activity you cannot observe any other way, and when getting the hire wrong is expensive.
Writing roles are the clearest case. Nobody can talk their way into being a good writer, and nobody can fake it in a two hour edit. Data roles are close behind. Sales roles are more debatable, because a live call simulation gives you most of the signal in twenty minutes.
Where it does not earn its place: senior leadership hires, where the actual work is judgement over months, and roles where you already have a portfolio, a public record or a strong reference network.
Length is the whole argument
Most of the resentment is not about being tested. It is about being asked for six hours of unpaid work by the third of five companies in a process.
Two hours is the practical ceiling for an unpaid exercise. If your task genuinely needs more, pay for it at a real rate. Paying also fixes the design, because once you are writing a cheque you get ruthless about scope.
Design rules that hold up
Give the real thing, cleaned up. A sanitised version of an actual problem beats an abstract puzzle, because it tells the candidate what the job is and lets them decide if they want it.
Say how it will be judged. A short rubric shared in advance does not reduce signal. It reduces the number of candidates who fail because they guessed wrong about priorities.
Never ask for work you will use. It is unethical, and in several jurisdictions it looks a lot like unpaid labour rather than assessment.
Review blind where you can. Names off, review before the debrief, two reviewers who score independently. Otherwise the exercise just launders the impression from the phone screen.
The compliance edge
In the US and Canada, unpaid work samples sit in a grey zone if they resemble real production tasks. In parts of Europe the line is sharper, and works councils in Germany and the Netherlands will ask questions if the exercise is long or unpaid.
Paying a flat fee for anything over two hours resolves most of this and costs less than a single bad hire.
The version most teams should run
One exercise, two hours or less, sent after the first interview rather than before, judged against a written rubric by two people who did not do the screen. Feedback given to everyone who completed it, whether or not they move on.
That last part costs an hour a week and does more for your reputation with candidates than any careers page.