Take-home Assignment

Measuring skill in an interview room has a ceiling. Producing an answer in forty minutes while three strangers watch has almost nothing in common with the job. Real work involves looking things up, revising something already built, and writing down why it was done that way. Strip all of that out and what remains visible is how practiced someone is at that particular format.

So the results come out crooked. Someone who reliably produces good work over a day gets rejected, and someone who has drilled interview puzzles passes. Cases where the two people's actual output runs the other way are not rare. The interview measured composure while being watched, which is a real skill and not the one most days of the job call for.

Résumés and portfolios don't solve it either. The work looks strong, but nowhere does it say which parts were this person's. A team artifact posted as a personal one and something genuinely finished alone look identical on the page.

A take-home assignment changes direction by building conditions closer to the job. Give a few days and the freedom to look things up, take the result, and read how the person judges. The cost is that it takes days of a candidate's time, so without a stated size, the number of people who abandon the application goes up.

Design the assignment, send it with the criteria, take the submission, score it with more than one reviewer, then confirm the reasoning in a follow-up conversation. Size the candidate's side at under four hours and give five to seven days of calendar time.

1. Design the assignment. Pick a problem lifted from real work but not something currently in flight. Handing over live work makes it unpaid labor rather than hiring. Real customer data stays out too. A problem from last quarter that already shipped works well, because the team already knows what a good answer to it looks like.
2. Send the criteria with it. What's being read, what format to submit in, and how many hours it was designed to take all go in one document. Hide the criteria and candidates spend time on things nobody is reading, while the reviewers turn that spent time into a deduction.
3. Ask for a short note with the result. Why this approach, what was deliberately left out, what they'd do with more time — three lines is enough. Those three lines often say more than the artifact.
4. Have at least two people score independently. Same rubric, no visibility into each other's numbers, then compare. With a single reviewer, the submission built the way that reviewer happens to work scores highest.
5. Book a thirty to forty-five minute follow-up. Open the submission together and ask why a particular choice, and what they'd change if a condition moved. Whether someone else built it usually surfaces here.

Two things get fixed in advance. One is the size cap. There is no way to verify how many hours a candidate actually spent, so the only lever against people who can spend more is making the assignment small. The other is telling rejected candidates the outcome. A stage that takes days of someone's time and answers with silence is a reason not to apply again.

Design and scoring sit with two people who actually do the job, and the hiring manager makes the call. Sending the assignment, tracking the deadline, and returning the result belong to whoever runs the process.

The placement is after screening, and after the technical interview rather than before it. Put it ahead of the interview and the pool is still large, so the scoring load multiplies, and asking for a multi-day assignment from someone who hasn't spoken with the company yet pushes the abandonment rate up. Put it after and only a few remain, at the cost of about a week added to time to hire. Which trade a team takes usually comes down to how many candidates are in flight at once.

It gets skipped in two situations: when public work is already sufficient to judge from, and when someone's career is long enough that asking people who worked with them is more accurate. The second case gets covered by a reference check instead.

Some companies pay for the assignment — an amount matching the four hours it was designed to take. Acknowledging that it takes someone's time lowers the abandonment rate noticeably. Roles get divided so the stages don't check the same thing twice: the artifact and the reasoning belong here, while how someone behaves when opinions split belongs to the culture interview.

"The assignment is designed for four hours, and the window is five days."

The size is fixed so that spending more time doesn't become the advantage.

"The three lines on why they did it that way helped more than the artifact did."

Confirming that this stage reads the reasoning.

"We can't hand out the screen we're currently building as an assignment."

Using live work makes it unpaid labor.

Pagedeck revised the assignment twice for one design opening. The first version asked candidates to redesign the product's whole screen flow, and three of the six who received it didn't submit inside the window. What Noa established was that the problem was size rather than skill: she worked through it herself and it took eight hours.

The second version shrank to finding one problem in an existing screen and proposing a single alternative. A three-line field for the reasoning went in, and the rubric went out with it. All six submitted. Noa and Theo scored separately, and for the one candidate whose scores diverged sharply, Mia read the submission as a third reviewer.

The most common reading is that a polished artifact means a pass. The submission is what's visible, so that's how it gets read. But there's no way to know who spent four hours and who spent twelve, and the artifact alone won't say whether anyone helped. What to read is what they chose to leave out and why. Two submissions of equal polish can represent four hours and an entire weekend, and only one of those is repeatable once the person is on the team. Left uncorrected, the advantage goes to whoever can spend the most time, and the offer goes out without anything having been learned about how fast the person actually works.

The other reading is that a bigger assignment supports a better judgment. More to look at sounds like more signal. But submissions drop as the assignment grows. People exploring a move are usually employed, so eight weekend hours is hard to find, and anyone comparing several companies drops the one asking for the largest assignment first. Left uncorrected, the candidates with the most options leave first, and the choice gets made among whoever remained.