A working session answers the one question no interview format answers well: what is it actually like to work with this person? For 60–90 minutes, the candidate and one interviewer work a realistic task together — thinking aloud, trading ideas, hitting a snag and getting past it. Run well, it is the highest-signal stage in a loop. Run badly, it is a performance exam with an audience, which measures stage nerves.

Choosing the task

• Realistic but self-contained: drawn from the kind of work {{Role title}} actually does, stripped of internal context a newcomer could not have. Ten minutes of context-setting maximum.
• Never live work: using a current business problem is free consulting and candidates can tell. It also makes evaluation incomparable across candidates.
• Sized for two-thirds of the slot: a strong candidate should reach a meaningful milestone with time to spare. Completion is not the goal; the collaboration is.
• Deliberately underspecified in one dimension — a missing requirement, an ambiguous constraint — so you can watch how they surface and resolve ambiguity.
• The same task for every candidate in the search, with the same prepared hints.

Setup

• Send logistics in advance: format, duration, tools, and what will be evaluated. Surprise is not signal.
• Let candidates use their normal environment and tools where practical — familiar setup measures ability, unfamiliar setup measures adaptation to your setup.
• Prepare a written hint ladder: three escalating hints for each place candidates typically get stuck. Scripted hints keep help consistent; improvised help varies with how much the interviewer likes the candidate.

Facilitating — you are a colleague, not a judge

1. Open (5 min): Frame it honestly — we'll work on this together; ask me anything; I care about how we work, not whether we finish.
2. Context (10 min): Present the task, answer questions. Note which questions they ask — scoping questions are early signal.
3. Work (40–60 min): Behave like a good teammate: answer questions directly, offer opinions when asked, push back once on one of their choices to see what they do with disagreement. Give hints from the ladder when stuck-time stops producing information — watching someone flounder for fifteen minutes teaches you nothing new after the third.
4. Wrap (10 min): Ask: where would you take this next, what would you do differently, how confident are you in what we built?

Evaluation rubric

Scoring cautions

• Score against the hint ladder record: a candidate who used two hints and collaborated brilliantly can outscore one who needed none and worked like a sealed box.
• Nerves in the first fifteen minutes are noise. Weight the back half of the session, once the person has settled into the work.
• Do not reward speed for its own sake — the session simulates a workday, not a game show.