The Reasonable Take-Home Standard, and What Fails It

You just got sent a take-home assignment and now you have to decide, tonight, whether to spend your evening on it. This guide is for candidates who want a fixed rubric instead of a gut feeling: a way to grade the assignment in front of you and pick one of four moves - do it in full, time-box it, ask for pay or an alternative, or decline - each with the exact message to send. It replaces a stressful judgement call with a scored checklist.

Most pages on this topic argue whether take-homes are ethical. This one does not argue. It gives you criteria two people would grade the same way, and the scripts that follow from the grade.

What counts as a reasonable take-home?

A take-home is reasonable when its true completion time sits at or under about two hours for most roles, when it does not hand you the company's live work to do for free, and when nothing you submit is silently assigned away as IP. Those three tests - time, real-work signal, and ownership - are the whole standard. Everything below makes each one gradeable.

The time ceiling is the load-bearing test, and it is not a comfort preference. A named DEI consulting firm recommends a 2-hour limit for nearly all roles, with less for lower-paid and entry-level roles, and 30 to 60 minutes for SDR positions. The reason is equity, not politeness: low-paid candidates apply to more jobs with fewer spare hours, so a long assignment filters on free time rather than skill. The same firm draws the pay line at that threshold. If the team insists on more than two hours, they should pay you for it.

Design-side guidance lands a little higher, calling three to six hours reasonable and anything past a day's work a red flag. Candidate communities cite a personal tolerance of three to five hours for a job they genuinely want, and read any ask beyond that as abuse. So the numbers cluster: two hours is the careful ceiling, three to five is the outer edge of tolerance, and a "weekend" fails outright.

The time thresholds, by source

The reasonable ceiling is not one number but a narrow band, and the pay-above line sits right at the bottom of it. Here is where each documented source draws the line, so you can see the consensus rather than trust a single opinion.

Read the table as a range, not a contradiction. The strictest source doubles as the pay line: two hours is both the reasonable ceiling and the point past which you should expect compensation. Use two hours as your default pass threshold and treat three to five as the borderline zone where time-boxing lives.

There is one number that already exists on the paid side. One firm pays final-round candidates for a 10-hour project. That matters because it proves paid assessments are a normal, workable practice, which is exactly why asking to be paid for a long take-home is not an outrageous request. It is the same trade the strict firms already recommend.

Why the stated estimate lies, and how to correct it

The single most common way this grading goes wrong is trusting the time estimate on the assignment. Candidates repeatedly report a "2-hour" prompt that actually took four to six quality hours, equivalent to a normal eight-hour work day. So the first correction is arithmetic: multiply the quote before you grade it.

The gap is structural, not accidental. Candidates who overrun do not admit it - someone who spent ten hours says four to six to avoid looking incompetent - so companies never learn their estimates are wrong and keep publishing them. You inherit an estimate that has been quietly deflated by everyone who came before you.

Two things routinely hide inside the quote. First, environment setup: one candidate spent five hours standing up a new environment and an hour on tutorials before touching the actual task. If the stack is unfamiliar, add that time separately rather than folding it into the estimate. Second, unstated rejection criteria - the instructions said unit tests were optional, and the rejection reason was a lack of unit tests. When "optional" turns out to be scored, the real scope is larger than the prompt admits, so budget for the optional items too.

Telling generic prompts from your unpaid free work

A take-home crosses from assessment into free labor when the prompt maps onto the company's live roadmap rather than a made-up scenario. That is the clearest documented signal - in one case the assignment was plainly asking the candidate to map out the next stage of the company's development. Treat that mapping as your real-work flag.

There are observable tells. A named design case listed them: un-timeboxed, high-fidelity work on an obviously real project, a full responsive design plus a working prototype, and complete taxonomy and selection logic that engineering would normally build. When an assignment wants production-grade deliverables emailed back, on a project that is visibly the company's actual product, the risk that your output gets used is real.

But the flag is risk, not guilt. Intent as a measured rate is not established publicly. Some companies keep prompts deliberately unrelated to their business and say so, precisely so applicants do not think they are asking for free work. A domain match is a reason to ask a direct question, not to accuse. Ask whether your output will be used, and watch how they answer.

The IP terms you have to read before you start

Whether you keep rights to what you make turns on one clause, and it is quiet. In many situations a contractor keeps copyright in what it creates unless the contract assigns it, so the assignment clause is what actually moves ownership. The danger is not the assignment itself. It is a signed work-product clause you skimmed past because "it's just an interview."

Work-product assignment clauses typically transfer ownership of all work product, inventions, or deliverables created during a project, across all materials, software, and IP regardless of medium. There is a practical test for whether you have signed your rights away: if a clause only licenses what you made rather than assigning it, you do not own it, whatever anyone says. Read for the words "assign" and "all work product," not for reassuring tone.

Practitioners protect themselves in three documented ways. Watermark any images or content you submit. Ask for a hypothetical, non-real-world scenario, so you can be sure you are not doing free work for a company that might not hire you. And submit framework-only where you can: one candidate offered to discuss the approach, how their experience aligned, and past problems solved, but not the finished solutions. Whether a take-home normally comes with a signed IP-assignment agreement is not established publicly, so treat every attached form as one to read rather than assume.

The procedure: from assignment to sent message

Run these eight steps in order. The first five produce a grade, the last three turn the grade into a sent message and a tracked outcome. The whole thing takes about an hour, most of which is the ten minutes you spend deciding rather than the evening you were about to lose.

One caveat on step one: sources disagree on when a take-home should appear in the process. Some teams send them early, while others insist they belong only very close to the final stage. An early-stage take-home for many hours of work is a worse trade than a late-stage one, because you are spending IP and time before the company has committed much to you. Weight the stage into your move.

The four moves, with copy-paste scripts

Your grade points to exactly one of four moves. Do the assignment, time-box a partial version, ask for pay or an alternative, or decline. Each has a documented script and a documented downside, so you send the right message instead of writing an anxious one from scratch.

Thanks for the assignment. I want to give this real attention without turning it into a full day, so I have time-boxed it to a focused effort and documented the rest.

Attached is my partial solution. Any images and content are watermarked. In the write-up I have listed the remaining work, the assumptions I made, and the trade-offs behind each decision. I would welcome the chance to walk you or the hiring manager through it live and cover what I would do with more time.

Option A, alternative: I am very interested in this role. Before I start, could you tell me what this assessment is meant to uncover? I may be able to demonstrate the same thing by walking through a past project or presenting a portfolio, which would let us cover more ground together.

Option B, pay: I would be glad to go over a scoped version of this on a live 30-minute Zoom at no cost. For work beyond that, I would ask to be compensated for my time, in line with the effort the full assignment requires.

Thank you for the opportunity. After reviewing the assignment, I am not able to take on a project of this scope as an unpaid exercise. I would still love to move forward. I can share a GitHub repository and proof of relevant work, and I am happy to walk through my past projects in a live conversation so you can assess my fit directly.

The move-one script is the short one: for a pass-plus-generic assignment, confirm the deadline, do the work inside your time box, and submit. No negotiation is needed when the trade is already fair.

Know the downsides. Asking for an alternative carries an explicit risk: be prepared to potentially be removed from the running if you do not participate. Declining does not always end candidacy, though - one candidate who sent a GitHub link and proof of work was still considered for the next round at two companies. And the time-box script works because a legitimate company will take the walkthrough discussion. A company that refuses to talk it through has told you something.

A domain-matched prompt flags risk, not guilt; the honest move is to ask whether your output will be used.

How this standard fails, and the false positives to catch

The most valuable part of any rubric is knowing when it lies to you. These are the documented ways a take-home passes or fails your grade for the wrong reason, and the check that catches each one.

Two of these deserve extra weight. First, completing the assignment does not guarantee anything - even paying does not fix the drop-out problem. Before one named company pivoted away from take-homes, 20% of candidates would simply not complete them, and pay did not always change that, which tells you the friction is time and IP risk, not just money. So do not treat submission as a contract for advancement; the common failure is being ghosted or rejected with no explanation after a week of work.

Second, "paid" is not automatically fair. A token stipend can be worse than none because it launders a large ask as reasonable. Grade the pay against the true-time figure you built in step two, the same way you grade the time itself.

What the talent pool says about your leverage

Your ability to decline or negotiate a take-home is partly a function of how scarce your profile is, and that scarcity is measurable. In Refolk's index of professional profiles, the size of your talent pool sets how replaceable you are to the team that just sent the assignment. A thin pool is leverage.

The pattern is worth reading before you decide. Senior and specialised profiles are scarcer, and scarcity is exactly what lets you ask for pay or propose a portfolio walkthrough without ending the conversation. If you are one of 5,557 US data scientists with Python or one of 4,288 UK Python engineers in the index, the company has fewer easy substitutes, and that changes the tone of a negotiation. Refolk writes your resume from your own history and scores how well you fit a posting, which tells you where you sit in that pool before you weigh a demand.

The same figures cut the other way for a very deep pool. US software engineers with Python number 55,946, so if that is exactly your profile, a decline is easier to replace and your negotiating room is narrower. Grade your leverage honestly, then let it inform whether you pick move one or move three.

Before you call the decision made

Run this checklist before you either start the work or hit send on a script. If any item is unchecked, you have not finished grading and you are about to react instead of decide.

Keep the standard current by feeding it your own outcomes. Every take-home you grade and every response you get - advanced, dropped, ghosted, or rejected - is a data point that sharpens your realism factor and your read on which companies negotiate in good faith. The two-hour ceiling and the IP tests are stable, but your personal 2x-to-5x multiplier gets more accurate the more take-homes you run through this rubric. Re-check the pay line the same way: if you keep being offered token stipends against large asks, that is a signal about the market segment you are in, not about your grading.

The point of the standard is not to make you refuse take-homes. It is to make the decision take ten minutes and a script instead of a lost evening and a knot in your stomach. Grade it, pick the move, send the message, and move on to the next posting.