Guide
A remote assessment is easy to send and easy to get wrong. The failure modes are rarely technical — they are a test that measures the wrong thing, or a signal read as more than it is.
Write down what the first three months of the role actually require, then test that. Screening rounds drift toward what is easy to auto-mark, which is how a role that needs judgement ends up filtered on trivia recall.
Keep it short. A screening assessment that takes ninety minutes selects for candidates with ninety spare minutes, which is not the trait you were hiring for. Thirty to forty-five minutes is usually enough to separate a shortlist.
Fair comparison needs the clock enforced server-side, not by the candidate’s browser; identical questions or a documented randomisation with the shown order recorded; consistent marking, which means one reviewer applying one standard to short-answer responses; and attempt limits set deliberately rather than left open.
Browser-level signals — focus loss, copy, paste, developer tools — are worth collecting and worth reading carefully. They cannot see a second device or another person in the room, so they are not proof of anything on their own.
Set a threshold high enough that ordinary behaviour does not trip it, review borderline cases as a human, and never let an automated flag end a candidacy by itself. Tell candidates what is collected before they start: it reduces accidental violations and it is what makes a later decision defensible.
Tell candidates what happens next and by when, then do that. Whether you share scores is a policy decision — on this platform scored results are only sent when you explicitly share them, so it is a choice rather than a default. Silence after an unpaid hour of someone’s time is the most common own goal in remote screening.
Long enough to separate a shortlist and short enough that availability is not the filter — usually thirty to forty-five minutes. Longer rounds select for candidates with spare time rather than for the skill you are testing.
Answers autosave and the candidate resumes the same attempt against the same server-anchored deadline. Reconnecting neither restarts the attempt nor extends the clock.
Not on the count alone. Browser signals cannot distinguish a deliberate search from a notification stealing focus. Use them to decide what to review, keep a human in the loop, and pair high-stakes decisions with a round where someone is present.
More answers on the full FAQ.