The Workbench gives candidates a real working environment — an editor, a terminal, runnable checks — instead of a whiteboard and a guess. You attach a challenge to an interview stage; every candidate in that stage gets the same one, which is what makes comparing them honest. Candidates can see every check the grader runs — there are no hidden test cases.
Challenges run two ways: as an async take-home the candidate does on their own time, or as a live session worked during the call. Either way, Interviews is where the results land.
Enabling a challenge on a stage
- Open Interviews → Challenges with a position selected in the sidebar. Each interview stage gets a row.
- Click Configure on a stage and flip Challenge on this stage on.
- Pick a kind (coding, architecture, code review, debugging) and either a preset scenario or a repo of your own — see below.
- Tell it what you want to see. A sentence is enough — "how they handle a messy legacy service", say.
- Hit Generate now. HQ drafts the brief, tasks, checks and rubric, then bakes the workspace image. The chip walks through Generating and Building to Ready — building takes a few minutes.
If generation fails, the chip says so and the sheet shows why, with a Retry. The preview tabs (Brief, Tasks & checks, Diagrams, Rubric) show exactly what the candidate will get — read the brief before you send anyone in.
Grounding a challenge in your repo
Preset scenarios are fine. Your actual codebase is better — the candidate works on code that looks like yours, and the challenge stops feeling like homework from a textbook.
Connect GitHub from Settings → Integrations (or the Connect GitHub button in the challenge sheet), install the app on the repos you're willing to use, then pick a repo in the sheet — paste owner/repo, a GitHub URL, or choose from the connected list. An optional branch/ref pins what gets snapshotted. Once a repo is set, the preset choice disappears; the repo is the scenario now.
If the button says it's not available on your environment, GitHub isn't configured for your deployment yet — presets still work.
Sending an async take-home
Take-homes ride the normal scheduling wizard. When the stage you pick has a ready challenge, a Format choice appears: live session or async take-home.
- Choose Async take-home. The interviewer and meeting-link steps disappear — there's no meeting.
- Set the complete-by date (default: a week out) and the time budget. The clock starts when the candidate opens the workspace, not when you send the email.
- Create the invites. You get a workspace link per candidate — copy it into your own email. The link is the invitation; Chosen doesn't email the candidate for you here.
If a link shows as still provisioning, give it a minute and copy it from the row on the Interviews page instead.
Reading the results
When a candidate submits, the scorecard shows a work session block on top: active time, how the code arrived (typed vs. pasted vs. AI-generated), prompt counts, check runs, tab-aways, and chapter markers for how the session unfolded. Async sessions arrive auto-scored — graded from the checks and the session evidence, marked with an amber badge. Your review supersedes it: score each criterion yourself and submit like any other scorecard. The auto-score is a first read, not a verdict.
Watch replay opens the full session — every keystroke, command, and AI exchange, replayable. Candidates know all of this is captured; it's shown to them before they start.
Live follow-ups
To see a strong take-home candidate think in real time, use Set up live follow-up on the session. It creates a follow-up interview seeded from their submitted work — schedule it like any other meeting and the candidate picks up where they left off, with you watching. Available once the session is submitted.