Priced per interview, because that is the unit of work.

No. A credit is spent when an interview is completed, not when it starts, so a session that ends because of a technical failure on our side is not charged. A silent failure in a hiring workflow costs a candidate a job, so we do not treat these quietly.

Nothing is consumed and the candidate is not penalized. They are routed to a request for a human interview and you are notified. Declining has to be a genuine option, otherwise the consent is not consent, and the whole compliance position falls apart. Roughly 5 to 8 percent of candidates decline, and you should plan for a human path.

One credit. Every interview costs the same: Screen, Coding, System Design or Data Science, at any duration. One credit, one interview, no table to memorize. Duration is set per interview type when you create the job and is fixed for every candidate on that job, so your scores stay comparable.

Three cases: Lead/Staff+ System Design, where scope-shaping and operational reasoning live in exactly the phases that get squeezed at 45; Coding where you want a real second problem rather than one stretched thin; and any job where this is the only interview you are running. Otherwise 45 is the calibrated default: the extra 15 minutes at 60 goes almost entirely to the deep dive, not to requirements gathering, and most jobs do not need it.

Yes, by setting the allowed languages to just that one, which means the candidate sees the constraint before they start. Separately you can set an optional preferred choice of language: the editor opens in it and the interviewer mentions it once, then explicitly tells the candidate to use whatever they are fastest in. That preference is never sent to the scoring engine, so it cannot reward a candidate for complying with it. You see which language they chose on the report as information, not as a score.

Plan interviews reset monthly and do not roll over. There is no overage charge: nothing appears on next month’s invoice for going over. Instead you get a warning in-app and two real choices, both priced against the plan you already pay for. Buy a pack (10, 25, 50 or 100, expiring 12 months after purchase, at 15% under your plan’s per-credit rate on Starter, Growth and Scale, or 5% under on Entry) for an immediate top-up, or upgrade if the volume looks durable rather than a one-month spike. A pack and the monthly allowance are separate balances; the monthly allowance is always used first, since it is the one that expires at the end of the month.

Never, on any plan, and this is a policy decision rather than a feature gap. Live produces a recommendation with evidence. A person at your company dispositions the candidate. When ATS write-back ships, which is planned and not built yet, it will require human confirmation by default.

No, not by us and not by any of our vendors. Zero-retention or no-training terms are a requirement in every AI vendor agreement we sign, and our sub-processor list is public. Aggregate, de-identified score statistics are used to improve the rubric, and that is disclosed to candidates in the notice.

Adjustable weights on Scale and Enterprise are planned, not built yet. When they ship you will be able to adjust the weighting between the four dimensions by up to 0.10 each. You will not be able to add, remove, or redefine the dimensions. The published rubric is what makes the bias audit and the defensibility argument work, and a per-customer rubric would quietly destroy both.

United States and Canada. Canada excludes Quebec, because the Charter of the French Language requires a French interview, French notices and a French rubric, and translating behavioral anchors is a recalibration exercise rather than a translation one.

The EU is blocked rather than served non-compliantly, because employment AI is high-risk under the EU AI Act and doing it properly is a program, not a config flag. That includes Ireland, which catches people out: English-speaking and full of technology employers, and still in the EU. EU support is on the roadmap with data residency.

Retention is set per account, not per plan, and it defaults to twelve months for candidate identity and session recordings. You can configure recordings down to 30 days, and an Illinois candidate can request destruction of their own recording within 30 days regardless of your account’s setting. Other categories carry their own defaults: work artifacts (code, canvas, notes) for job close plus 24 months, evaluations and evidence for 3 years to support a bias audit or a dispute, and the audit log for 7 years, which is never shortened. The full table, and how deletion actually works, is on the security page.

Reports and artifacts stay available in read-only mode for 30 days, then everything is deleted on your retention schedule. You can export everything (reports, transcripts, artifacts, and the bias audit dataset) at any time, including after cancellation begins.