A Take-Home Test Needs a Hard Stop

Field guide — 10 November 2026

A task labeled “two hours” isn’t a two-hour assessment unless the brief makes the stopping point and success bar real. In a Hacker News discussion, one commenter said a task described as two hours consumed a whole weekend; another wrote that a five-working-day chatbot-backend assignment took 80–100 hours including the weekend. These are personal accounts, not representative evidence. They still expose a design risk: when the expected finish is implicit, candidates cannot tell whether honoring the suggested cap will count against them. S1

Define the contract before sending

Specify one core behavior, acceptance criteria, a hard time budget, and an explicit stop rule: submit what exists when time ends and list the unfinished work. Label extras as optional and unscored. A compact rubric can separate requirement interpretation, correctness against the stated behavior, tests or edge cases, and explanation. Y Combinator’s first-engineer guide recommends listing technical and nontechnical criteria and evaluating every candidate against them; the practical extension is to make the take-home’s criteria visible rather than letting reviewers infer them afterward. S2

Don’t silently promote polish or extra features into hidden requirements. If they matter, name them in advance and score them consistently.

Allow AI without making it a guessing game

If AI use is allowed, name the permitted tools, whether candidates may send assignment material to hosted models, and what they must be able to explain. Canva wrote in June 2025 that it expects backend, machine-learning, and frontend candidates to use AI tools during technical interviews and tells them in advance. That’s one company’s policy, not a universal standard, but the transparency principle travels. S3

Then observe judgment rather than asking candidates to prove prompt fluency. HackerRank’s July 2026 vendor guide suggests a multi-step build with a changed requirement halfway through, then watching how the candidate redirects the tool. Treat that as an interview-design suggestion, not evidence that this format predicts on-the-job performance. S4

Keep access, task, time limit, and scoring consistent across candidates. In a follow-up, ask what they verified, what they left out, and why. A candidate who identifies uncertainty and stops within the agreed limit has met the contract; extra unpaid hours should not become an invisible bonus criterion.