The Take-Home Nobody Finishes

You sent the assignment to six candidates and two sent something back. The problem usually isn't candidate interest — it's a brief with no ceiling, no rubric, and no return date. Here's how to build a take-home people actually complete.

By JobPost Team · Sep 20, 2026 · 4 min read

You sent the take-home to six candidates. Four said yes. Two sent something back, and one of those was nine days late with an apology in the first line of the README. You now have a stage in your process that costs you two thirds of the people who cleared the phone screen, and you are grading whoever survived it.

That is not a signal about candidate motivation. It is a signal about the brief.

Where People Actually Drop Out

There are three distinct exits, and they have different causes.

The first is at the moment of reading. The brief arrives, it is eleven hundred words long, it mentions a dataset and a deployment and "feel free to add anything you think demonstrates your skills," and the candidate closes the tab. They have not decided against you. They have deferred, and deferral is how most take-homes die.

The second is the weekend that never comes. They blocked Saturday. Something happened. The next open Saturday is eleven days out, by which point two other processes have moved to final round and yours has gone quiet.

The third is the worst one for you: they start, discover the thing is twice as large as advertised, and abandon it half-built. That candidate now tells people your company wasted six hours of their time. They are right.

"About Four Hours" Is Not A Constraint

A time estimate is not scope. It is a hope about scope, and candidates know it.

Put yourself on their side of it. You have a brief with no stated rubric, competing against an unknown number of strangers, for a job you want. How much effort is correct? There is no way to answer that except by doing more. So the person who respects your four-hour guidance turns in something honest and thin, and the person who spent the whole weekend turns in tests, a README, a deployed demo, and a section on trade-offs. You hire the second one. Everyone watching learns the lesson.

The fix is structural, not rhetorical. Adding "we really do mean four hours!" in bold changes nothing. What changes behaviour:

  • A scaffold. Send a repo with the boilerplate, the fixtures, and the build already working. Every hour they spend on setup is an hour you learn nothing from.
  • An out-of-scope list. Write down what you do not want: no auth, no deployment, no test coverage beyond the core function, no styling. Candidates will not believe the first one. They will believe a list of five.
  • One real decision. A good take-home contains a single interesting judgement call and the plumbing required to reach it. If your brief has three, cut two.
  • The rubric, in the brief. Tell them exactly what you are scoring. If you are not comfortable publishing it, you probably do not have one, and you are grading on vibes.

The Filter You Didn't Mean To Build

Ask who completes an eight-hour assignment on a 48-hour turnaround.

Someone whose current job is slack enough to absorb it. Someone without a kid, or with a partner who takes the weekend. Someone not already three rounds deep at two other companies. Someone who is not, right now, doing contract work that actually pays.

None of those are correlated with being good at the job. You built an availability test and you are reading it as a skill test. The strongest mid-career candidate in your pipeline — currently employed, two children, a commute — is the most likely person to quietly go dark on you, and you will record that as lack of interest.

So: give a week to return it, not two days. Let people ask for two weeks without penalty. And if the work is substantial enough that you feel uneasy asking for it for free, pay for it — a flat fee, stated up front, paid on submission whether or not you advance the candidate. In Canada, that is also the cleanest way to stay on the right side of the question of whether you have received work product for your own benefit. Nobody has ever been annoyed by a $300 e-transfer.

What You Owe After Submission

This is where most processes fail their candidates outright. Somebody spent a Saturday on your problem. The minimum response is a named reviewer, a date by which they will hear back, and two or three specific sentences about the submission itself — not a template rejection that could have been sent before you opened the file.

If you cannot commit to reading every submission within a week, you do not have the review capacity to run a take-home stage. Run a live exercise instead. Ninety minutes on a call with a real engineer costs the candidate less and tells you more, because you get to watch the reasoning rather than infer it from a finished artifact someone may or may not have had help with.

If You're On The Receiving End

Before you start, ask three questions: how long should this take, who reads it, and what is the decision timeline after I submit. All three are reasonable. The answers are diagnostic.

Then timebox it yourself and say so. Set a timer, stop when it goes, and open the README with: *"Timeboxed to three hours. Here is what I built, here is what I left out and why, here is what I would do next."* A hiring team worth joining will read that as judgement. A team that penalises it has told you something about how they will treat your time for the next three years, and the take-home has done its job in the other direction.