The Take-Home Invitation, Scored to Do, Cap, or Decline

A take-home invitation lands in your inbox and the clock starts before you have decided anything. Most published guides assume you have already said yes and coach the build; this one sits one step upstream, at the fork where you choose to do it in full, cap and negotiate its scope, or decline without killing the process. It is for candidates mid-loop who need a defensible answer today, not a philosophy of unpaid work. What follows is a scored rubric: five dimensions, three actions, and the exact reply for each.

What this framework decides

This framework turns one take-home invitation into one of three actions with a matching reply already written. You score the invite on five dimensions, map the scores to Do, Cap, or Decline, and send the reply that fits.

The three actions are not "yes" and "no." They are:

• Do - complete the assignment as scoped, confirm and ask one clarifying question.
• Cap - agree in principle, state a time box, offer a live walkthrough for anything beyond it.
• Decline - thank them, offer a substitute, keep the door open.

The reason a bare yes/no fails is that the invite itself is often ambiguous. A task close to the live product can be genuine signal or free work. A "two-hour" exercise can cost five once you count environment setup. The score exists to resolve that ambiguity into an action you can defend to yourself and to the recruiter.

That gap is the whole reason to score rather than reflexively comply. Companies fan assignments out cheaply, and each candidate pays full price. Capping effort beats maximizing it because the expected return on a marginal hour is low.

The five dimensions you score

Score every invite on the same five dimensions, in this order: realistic hours, free-work risk, stage and defense, evaluation criteria, and your market leverage. Each dimension proves something specific, and each has a way it lies to you.

Realistic hours, not stated hours

What it proves: whether the exercise is inside the norms practitioners accept. What it looks like when it lies: the stated number is almost always low. Assignment authors consistently underestimate their own exercises, so double the estimate before you judge it. One candidate spent five or more hours just setting up an unfamiliar environment for a task billed at one to two hours, then canceled.

The practitioner consensus is a three-hour ceiling for an unpaid exercise, with most of the signal already present by hour two. A HireVue engineering director put the outer bound plainly: assignments that take two days are too much to ask.

Free-work risk

What it proves: whether the output is a check or a deliverable. What it looks like when it lies: product relevance masquerades as legitimacy. A task can be close to the live product because that is what the team does all day, or because they want free labor. You resolve it by asking whether the output will be used and whether it is a toy slice or a real thing.

Named red flags from the field: the task is tied too closely to the company's actual product, there are multiple revision rounds, the output is publishable or deliverable such as real marketing campaigns or client-specific problems, and the role is constantly open. A product leader's counterpoint keeps this honest: for product roles, unless the assignment is truly badly designed, candidates should not spend more than a few hours, and the output will necessarily be lacking and not usable by the company.

Stage and defense

What it proves: whether a human is invested and whether your judgment will be examined. What it looks like when it lies: a polished-looking process with no debrief scheduled. The single cleanest free-work tell is a missing defense session. If no one plans to discuss your work, the work itself, not your reasoning, is the deliverable they wanted.

Evaluation criteria

What it proves: how graders will read the work, which tells you what to optimize. What it looks like when it lies: purposefully vague instructions, which some teams use deliberately to see how you think. Vagueness is not automatically bad, but paired with long hours it is a negotiate trigger.

Structured rubrics dominate real grading. Slack grades its exercise against more than 30 predetermined criteria to limit bias, looking for clean, readable, performant, maintainable output. Reviewers also read commit history: breaking work into logical commits helps the reviewer follow your line of thinking. This is why a scoped, explainable three-hour slice can outscore an over-engineered build.

Market leverage

What it proves: how hard you can push without losing the process. What it looks like when it lies: you overestimate it. Leverage is real only with competing options. Competing offers or scarce skills raise it; nothing lined up lowers it. With nothing lined up, negotiating aggressively can cost you the process.

Hours mapped to action

Hours are the fastest cut. Map the doubled estimate against the thresholds below before you weigh anything softer. This table is the spine of the Do/Cap/Decline call.

Note what doubling does here. A task billed at four hours becomes eight once you correct for author optimism, which pushes it from the reasonable band toward the decline line. The estimate you score is never the estimate you were quoted.

The adoption numbers that set your expectations

Take-homes are common and their pass rates are low, so treat each one as a bet with a known bad payout, not a formality you owe the company. The benchmarks below explain why capping beats maximizing.

The 41% hybrid figure matters for your defense-session check. Pairing the take-home with a live review is now common enough that its absence is a real signal, not just bad luck. When a review is scheduled, your explainable slice is what carries the round.

Companies fan assignments out cheaply and each candidate pays full price, which is why capping beats maximizing.

The procedure, invite to reply

Run these steps in order the moment the invite lands. The whole pass takes about ninety minutes of thinking and produces one action plus one sent reply. Sources disagree on whether to assess free-work risk or stage first; either order works as long as you complete both before you decide.

Mapping scores to the three actions

Once every dimension has a label, the mapping is mechanical. The matrix below reads free-work risk against leverage, which are the two variables that most often flip a borderline case.

The clean cases resolve on hours and defense alone. Doubled hours at or under three, after a human conversation, with a scheduled debrief, is a Do. Doubled hours between four and eight, or a vague brief, or no prior conversation, is a Cap. High free-work risk with a missing defense session and any real leverage is a Decline.

The messy cases are where leverage does the work. High risk with nothing else lined up means you cap and insist on a review call rather than refuse outright, because a flat refusal can end a process you cannot yet replace.

The reply scripts

Each action has a reply that keeps the process alive. Send one of these three, adapted to your specifics. The point of scripting them is that the hard thinking is already done by the time you write, so you do not soften a Cap into a Do out of nerves.

Thanks for sending this over - happy to take it on. Before I start, one quick clarification: [the single most scope-relevant question, e.g. "should I prioritize a working slice over full coverage if I hit the time box?"]. I'll aim to keep it within the estimated time and will note any tradeoffs I make so we can talk them through on the review call.

I'd like to move forward on this. To keep it fair to both sides, I'll time-box the build to [N] hours and hand over a working slice with a short note on the tradeoffs and what I'd do with more time. I'd also be glad to walk through it live on a call so you can probe my reasoning directly - any work beyond the time box I'd want to scope together first. Does that work?

Thanks for the invitation. Rather than a new build, I'd like to offer a faster way to see the same signal: I can walk you through [a comparable past project / a portfolio piece / public work at LINK] on a live call and answer any questions about my approach and tradeoffs. That should give you a strong read on how I work. Would that be a workable substitute here?

Two phrasings from the field are worth keeping in these scripts. Offering to go over the work on a live call, with anything past thirty minutes compensated, advanced processes. Getting the recruiter on the phone to talk through the project, rather than emailing back and forth, also advanced them. Naming a genuine competing offer to ask to expedite or skip a step is a legitimate move when you actually hold one.

How this goes wrong

The framework fails in predictable ways, and the failures are more costly than the occasional lost opportunity. Each mode below has a check that catches it before it costs you.

• Trusting the stated hours. A "2-hour" task can need unfamiliar tooling setup; one candidate lost five or more hours before canceling. Check: scope the setup and environment before starting the clock.
• Mistaking product-relevance for legitimacy. A task close to the live product can be genuine signal or free work. Check: ask whether the output will be used and whether it is a toy slice or a real deliverable.
• Over-delivering to win. You pass the screen but signal that you ignore scope, and rejections happen for "not enough effort" even at a cap, because graders weight judgment, not volume. Check: cap openly and defend the cap, do not hide it.
• Confronting the recruiter as a decline tactic. Calling the practice discriminatory can end the process cleanly rather than negotiate it. One candidate did exactly this and the process ended. Check: decline with a substitute, never with an accusation.
• Declining with no substitute. A bare no reads as disinterest. Sources stress offering an alternative and not ghosting. Check: every decline carries a portfolio, past-project, or public-work offer.
• Assuming a defense session exists. No debrief scheduled means padded or AI-assisted work goes unchecked, and that a genuine skills-check may not be happening at all. Check: ask if there is a review call.
• Ignoring leverage. With nothing lined up, negotiating aggressively can cost the process; leverage is real only with competing options. Check: read your leverage honestly before you push.

Leverage is a local number

Your ability to decline or negotiate depends on how thin the market is around you, and that varies enormously by geography. In Refolk's index of professional profiles, the screening-stage recruiter population differs by nearly two orders of magnitude across markets.

The mechanism is simple. In a thin market a single decline is more visible and harder to replace, so a substitution offer preserves optionality where a flat refusal would burn it. In a deep market you have more room to walk. In Refolk's index, the top US employers of these screening-stage recruiters include Snowflake, Blue Origin, and Google; in the German sample, Google and AWS lead. Knowing who owns the stage at your target helps you judge how the take-home will actually be graded, since Greenhouse recommends the hiring manager for the role create the take-home test, and the same manager often runs the defense.

If you want to pressure-test your own leverage, look at who actually issues and grades these assignments in your field before you reply. Refolk writes your resume from your history, tailors it to each posting, and scores your fit, so you walk into the take-home fork already knowing which processes are worth the hours.

Before you hit send

Run this checklist against your reply before it leaves your outbox. It catches the failures above and confirms your action matches your score.

Keeping this current

The thresholds move, so re-check the mechanism rather than memorizing a number. Take-home adoption was reported at 68% of companies with a hybrid live-review pairing at 41%, and both figures trend upward year over year; when you re-run this, look for the current adoption and hybrid rates from a fresh coding-platform report rather than trusting last cycle's numbers. The pass-after-completion rate near 20% and the completion rate near 30% are the two figures that justify capping, so watch whether they shift. Everything else in this framework is structural: double the estimate, find the defense session, read your leverage locally, and let the score, not the invite's tone, choose your action.