NeuroFrame and HireVue

Side by side

HireVue, Inc. — the vendor discussed on this page. HireVue is a trademark of HireVue, Inc.; every other product and company name used here belongs to its own owner. All of them are used nominatively, to name the products under discussion, and no logo or trade dress is reproduced. This page is published by NeuroFrame, which has no affiliation with HireVue, Inc.

What HireVue is

HireVue, Inc. is an American predictive-hiring platform: algorithmically scored video interviews, game-based cognitive assessments, job simulations, technical and language checks. Since 15 October 2019 a controlling stake has been held by funds managed by The Carlyle Group; the investment came out of Carlyle Partners VII, with Granite Ventures, Sequoia, TCV and management retaining minority positions. Primary sources give the company's location two ways, and both are live: EPIC's 2019 complaint to the FTC lists an address in South Jordan, Utah, while the Carlyle release and the author affiliations on the company's peer-reviewed paper say Salt Lake City.

The game-based part of the product arrived by acquisition. On 10 May 2018 MindX, founded in 2014, announced it had signed a definitive agreement to be acquired by HireVue; the same post names Mercer and Onfido among its customers. On 3 May 2023 HireVue closed its purchase of Modern Hire from the Riverside Company — per the deal's adviser, Modern Hire was at that point "trusted by more than 450 leading global enterprises and nearly half the Fortune 100".

Product news over the last twelve months. Assessment Builder launched on 19 February 2026; on 10 March 2026 the vendor announced the acquisition of Hireguide technology for voice-based interview agents, and named Jeremy Friedman as CEO. As of 4 August 2026 the game-based assessments product page is live, and in the 2025–2026 product announcements we read we found no statement retiring the game module.

Scale, as the vendor states it in its February and March 2026 releases: over 1,150 customers, over 60% of the Fortune 100, over 180 million assessments, 80 million video interviews and 200 million chat-based candidate engagements. This is the vendor's own account and has not been independently verified. For a sense of trajectory rather than a judgement: SHRM, reporting the withdrawal of facial analysis, wrote that the platform "has hosted more than 19 million video interviews for over 700 customers worldwide".

One more fact matters for understanding what the product is today. HireVue publicly withdrew visual and facial analysis. SHRM reports the feature was discontinued in March 2020; the vendor's own blog of 12 January 2021 frames it as a decision "to not use any visual analysis in our pre-hire algorithms going forward". The reason the vendor gave was that "our algorithms do not see significant additional predictive power when non-verbal data is added to language data": around 0.25% of predictive power in most cases, up to 4% for roles with heavy customer interaction, according to the company's Chief Data Scientist as quoted by Fortune on 19 January 2021. If you are reading older commentary about this vendor, check its date before acting on it.

How the assessment is built: format, tasks, duration

The short answer: the game block is a module inside an assessment platform rather than a standalone product. Figure 3 of the vendor's own "HireVue's Assessment Science" whitepaper shows the typical scenario as a combined session — short games plus a video interview, optionally with coding or a job simulation, in one flow. That framing matters, because the predictive-validity coefficients most often quoted about this vendor sit in Table 3 of the same whitepaper, where the table itself labels them as belonging to the interview assessments rather than to the games.

Duration. Published figures differ across vendor materials, so it is worth asking which package a given number describes. The candidate-facing page says each game takes approximately 3 minutes, that "some game-based assessments last only 7 minutes, others may last 15-20", and advises setting aside around 30 minutes of distraction-free time. The product page as of 4 August 2026 says the games are "each under 20 minutes". The vendor ebook names five pre-built cognitive packages and describes three of them: Cognition Short at 7 minutes, Cognition Standard at 11, Cognition Comprehensive at 14. One caveat on that last document: its content is dated 2020 — every page carries the running head "THE 2020 CANDIDATE EXPERIENCE PLAYBOOK" — and the 2025/07 in the file's URL is a re-upload date, not the date of the data.

Number of games. On the pages and in the two vendor PDFs we checked on 4 August 2026 we did not find a total count of games in the catalogue; the vendor's phrasing is "five pre-built game-based cognitive assessment packages" — packages, not games — and "a short series of engaging games". Four games are described by name in the open peer-reviewed literature, with their durations: Numerosity 3 minutes, Shapedance 3.3, Pathfinder 5, Singularity 3. Figures such as 13, 17 or 20+ circulate on third-party prep sites; we found no vendor confirmation of them.

What is measured. The product page names three domains: Personality & Work Style, Working with People, Working with Information. The 2021 whitepaper is broader — cognitive ability, emotional intelligence, the Big Five, and a set of work competencies including Adaptability, Dependability, Communication, Team Orientation and Drive for Results & Initiative — with the caveat that these competencies are assessed across the platform, not by games alone.

Two terms, because the whole comparison turns on them. A battery of mini-games is a format where the assessment is assembled from several short, self-contained tasks, each with its own rule and its own name. A single long simulation is a format where all behaviour is observed inside one continuous scenario with delayed consequences. By the published evidence HireVue sits closer to the first and is sold in combination with interviews; NeuroFrame is the second. Neither format is inherently superior — they differ in what they can observe and in what they cost a candidate in time.

What the vendor publishes on validity and reliability

The short answer: HireVue publishes both vendor-authored and peer-reviewed evidence, and the evidence about the game module that we found published is predominantly convergent — how closely the game score tracks the score of a conventional test.

The whitepaper "HireVue's Assessment Science" (October 2021, authors Leutner, Liff, Zuloaga, Mondragon) gives, in Table 2, coefficients of .51 to .67 for game-based cognitive assessments against the legacy ICAR test, with modelling samples of 364–647 depending on the combination of games. The column heading is "Multiple R", and a note specifies that it is based on out-of-sample model performance under stratified k-fold cross-validation — so this is the multiple correlation of a machine-learning model, not a pairwise r, and the distinction is worth carrying into the conversation. The same document reports .36–.45 for emotional-intelligence games against GERT-S and STEM, and .52–.72 for game-based personality assessments against the IPIP inventory.

The peer-reviewed publication we found is Leutner, Codreanu, Brink and Bitsakis in Frontiers in Psychology, 18 January 2023. It reports convergent validity with ICAR at r = 0.5 (95% CI by Fisher transformation 0.43–0.56) and test–retest reliability of r = 0.68 across 102 participants within four months. Study samples: 565 panellists plus 3,044 applicants, then 3,107, then 85, then 4,778. Three of the four authors held a HireVue affiliation at publication, which the paper states.

The authors state the limitation themselves. They write that the impact of bias penalisation and gamified format changes on predictive validity for job performance is "unclear", call this "a clear gap in the literature", and explain the mechanism: by losing convergent validity with traditional tests that have high criterion validity, game-based assessments might produce a weaker link with job performance. As of 4 August 2026 we found no more recent peer-reviewed criterion-validity data for HireVue's games in open sources.

One line to keep separate. The predictive-validity coefficients the vendor publishes — r = .25 to .49 against work outcomes, AUC .68–.81, on samples up to 53,194 flight attendants — sit in Table 3 of the same whitepaper and belong to the interview assessments. The table labels them as such. Candidate-experience figures (95% completion and an NPS of 70 in the 2021 whitepaper, "over 90%" completion in the ebook) and the marketing case numbers on product pages (£1M annual cost savings, 90% faster hiring, 16% increase in DEI metrics) are stated by the vendor; on the pages we checked on 4 August 2026 we did not find the underlying methodology alongside them, so we report them as vendor claims.

Audits, complaints and proceedings

The short answer: HireVue publishes external bias audits, and among them is an audit of the game module itself — a fact worth stating plainly, because the opposite is often assumed.

Under NYC Local Law 144, HireVue engaged DCI Consulting Group; the press release of 28 January 2023 states that "competency-based and game-based algorithms will be audited for bias with respect to race, gender and the intersectional combination of race and gender across multiple job levels and use cases". In the summary produced on 5 July 2023, covering applicants scored between 1 January 2021 and 31 December 2022, there is a standalone section headed "Think – Shapedance, Numerosity Aggregate Analysis" — the game-based instrument audited separately from the interview competencies. Its impact ratios run from 0.86 to 1.00: for gender, Male n = 15,099 at a selection rate of 0.68 as comparator and Female n = 7,504 at 0.66 for an impact ratio of 0.96; by race, the comparator group is Two or More Races (n = 758, rate 0.70); Asian n = 5,636 at a rate of 0.70 gives an impact ratio of 1.00, and White n = 11,345 at a rate of 0.65 gives 0.93; in the intersectional table the comparator is Female Asian (n = 1,801, rate 0.71) and the lowest figure is Female White, n = 3,737, rate 0.61, impact ratio 0.86. Selection rates are reproduced as the report prints them, to two decimals, and the report warns about its own arithmetic: "the aggregated impact ratios reported in the tables cannot be computed directly from the aggregated selection rates appearing in the tables". The section covers intern and new-college-graduate roles. A later summary was produced on 5 July 2024 covering 2022–2024 data.

Other sections of the same report cover competency models that include an interview component and therefore fall outside the game module compared here. In the summary produced on 5 July 2024 the auditor states that "[a] number of the requirements specific to NYC Local Law 144 are not aligned to contemporary adverse impact analysis practices", and explains ratios above 1.0 by its own aggregation method. We reproduce the game-module figures only, and draw no conclusion from them.

Two earlier reviews. ORCAA (O'Neil Risk Consulting & Algorithmic Auditing) audited in April and May 2020, with a report dated 15 December 2020, covering pre-built assessments for early-career and campus hiring across eight competencies that include both video responses and "performance on psychometric games"; the conclusion was that the assessments "work as advertised with regard to fairness and bias issues", and the report itself states "This was not a comprehensive audit of HireVue's use of algorithms", with custom assessments outside its scope. Separately, on 7 April 2021 the vendor published results of a scientific audit by Dr Richard Landers of Landers Workforce Science and the University of Minnesota — around 1,000 pages of documentation and 18 hours of meetings, covering job analysis, OnDemand interviews and game-based assessments. That audit was commissioned and published by the vendor.

Complaints and proceedings, stated only as the fact of filing. On 6 November 2019 EPIC filed a complaint with the US Federal Trade Commission asking it to investigate the company's practices under Section 5 of the FTC Act; that is the complainant's position, and we assert nothing about any regulatory outcome. In Illinois, a settlement is pending in the class action Deyerler, et al. v. HireVue, Inc. (case no. 2026LA00000141, Circuit Court of Lake County) under the state's Biometric Information Privacy Act; the class covers people in Illinois who took an interview on the platform between 27 January 2017 and 25 June 2026 that involved a model that may have collected voice and facial biometrics, with a final hearing set for 28 October 2026. On the settlement site as of 4 August 2026 we found no fund amount published, and the settlement contains no admission of liability.

Context on the audit regime itself, which applies to the whole market rather than to any one vendor. The FAccT 2025 paper "Auditing the Audits" analysed 44 reports containing 116 bias audits published between July 2023 and early November 2024. Across that corpus, 53% of audits contain at least one impact ratio below 0.8 and 83% report missing demographic data. The authors also observed repeated identical results across employers' reports produced by DCI Consulting, most of which described HireVue tools — and they state directly that they could not identify the cause, offering a lawful explanation: Local Law 144 permits pooling data from multiple employers, so audits of the same tool by the same auditor may legitimately share results. We reproduce that caveat because without it an observation turns into an accusation.

What the law actually requires — as of 4 August 2026

The short answer: no jurisdiction we reviewed bans automated or game-based assessment. What they require is explainability, documented bias testing, and a right for a human to intervene. That is the frame in which both products should be evaluated.

New York. Under the DCWP rules implementing Local Law 144 (6 RCNY §§ 5-300 to 5-304, adopted 6 April 2023), the duty to commission an annual bias audit and publish its summary falls on the employer or employment agency — not on the assessment vendor. A vendor that commissions and publishes its own audit is therefore doing more than the rules require of a vendor; whether such an audit also satisfies a particular employer's own obligation is a question for that employer's counsel. The rules also define independence tightly: an auditor is not independent if it participated in developing or distributing the tool, is employed by the employer or the vendor, or holds a direct or material indirect financial interest in either. Separately, the law triggers not on the presence of AI but on the weight of its output: relying on a simplified output alone, weighting it above any other criterion, or using it to override conclusions reached from other factors including a human decision.

European Union. Annex III, point 4(a) of the AI Act names systems intended for the recruitment or selection of natural persons, including evaluating candidates, as high-risk. The application date for those Chapter III obligations, however, moved: Regulation (EU) 2026/1744 of 8 July 2026, published in the Official Journal on 24 July 2026, sets 2 December 2027 for Annex III high-risk systems and 2 August 2028 for Annex I. Anything written before July 2026 will still name 2 August 2026 as the date, because that is what the Act said until the amendment; check the date on any material you are shown before relying on it. The deferral did not touch the prohibitions: Article 5(1)(f), which bars AI systems inferring emotions of a natural person in the workplace, has applied since 2 February 2025, and the Article 50 transparency obligations remain at 2 August 2026.

GDPR. Article 22 gives a person the right not to be subject to a decision based solely on automated processing that produces legal effects or similarly significantly affects them, with three exceptions, and where those apply the controller must provide at minimum the right to obtain human intervention, express a point of view and contest the decision. Two CJEU rulings sharpen this for assessment vendors specifically: SCHUFA (C-634/21, 7 December 2023) held that a scoring provider's own automated production of a probability value can itself be an automated individual decision where the recipient strongly draws on it; and Dun & Bradstreet Austria (C-203/22, 27 February 2025) held that trade secrecy does not, as a rule, excuse a controller from giving an explanation sufficient for the person to contest the decision.

The Gulf. There is no dedicated AI-in-hiring statute in the UAE. The binding provision is Article 18 of the federal Personal Data Protection Law, which gives a data subject the right to object to decisions issued through automated processing that produce legal effects or seriously affect them, including profiling, with exceptions for contractual necessity, other legislation, or prior consent; Article 21 requires a data protection impact assessment where processing involves systematic and comprehensive evaluation of personal aspects based on automated processing. Within the DIFC, Regulation 10 to the Data Protection Law goes further, requiring advance notice that an autonomous system is in use, documentation of bias-detection and human-intervention mechanisms, and, for high-risk processing, certification and a designated officer. Across the wider MENA region, personal data protection laws are the binding foundation; national AI ethics charters are advisory.

MENA and language

The short answer: HireVue's presence in the region is confirmed and enterprise-scale; Arabic localisation is not something we were able to confirm from vendor sources, and it is the single most useful question to put to any vendor selling assessment into the Gulf.

On 4 December 2024 the vendor announced a partnership with Emirates NBD in Dubai. The release states the rollout "will initially target high-volume roles in retail, customer service and sales, with plans to extend to specialised positions in Technology, Capital and Treasury", and notes the bank receives up to 10,000 applications for a single role — a volume at which a 7-to-14-minute screening step and a 30-to-60-minute one are genuinely different operational propositions. In that release we found neither a list of the products deployed nor any mention of Arabic. A customer story for Emirates Group is hosted on the vendor's own site.

On language, the vendor's own whitepaper is the most precise source available: assessment content — instructions, test items and interview questions — is localised "using experts in psychometric test translation", local norms are used by country and region, "all models that rely on verbal behavior are language-specific", and cognitive game-based assessments "make minimal use of language or prior knowledge". That last point is a statement about construction rather than marketing, and it is mechanical: a task that leans on spatial reasoning and timing carries across languages more easily than one built on words. On the vendor pages we checked on 4 August 2026 we found neither a stated number of supported languages nor confirmation of an Arabic interface for the games; that is a statement about our search, not about the vendor's product.

Our own position here is worse, and we would rather you hear it from us. NeuroFrame delivers the assessment and the report in Russian and English. The client-facing library of role benchmarks — the 2,296 records that make the output interpretable — exists only in Russian. There is no Arabic. If your candidate flow is Arabic-speaking, this is a live constraint for us today, not a roadmap item we can wave at.

Norming is the second half of the same question. Both HireVue bias audit summaries we opened — 5 July 2023 and 5 July 2024 — were produced under US rules (NYC Local Law 144). On language and norming the vendor's own whitepaper says the opposite of a single-market model: "To ensure that candidates are only compared to others in their country or region, local norming groups are implemented". What we did not find on the pages we checked on 4 August 2026 is published norming data for Arabic-speaking samples. Our comparison sample of 14,850 people is drawn from 500 companies across 21 industries, 21 functions and 8 grades, and is likewise not a Gulf-normed sample. Anyone selling you a Gulf-normed instrument should be asked to show the sample composition, not the claim.

Where the NeuroFrame approach differs

The short answer: we observe one continuous scenario for 30–60 minutes and read eight parameters out of it — three cognitive (Mental Efficiency, Learning Agility, Progress Monitoring) and five personality (Openness to New, Risk Appetite, Result Focus, Agreeableness, Conscientiousness). We are not claiming this is more accurate than a battery. We are describing what it makes visible.

A continuous simulation with delayed consequences is designed to observe complex problem solving — a construct that overlaps with general cognitive ability by only about 18% of variance (r ≈ .43 in a meta-analysis of 47 studies). That overlap figure is the point: a substantial part of what the construct captures is not what a classic ability test captures. A second observable that a long session affords is behaviour under sustained load — vigilance decrement, the decline in sustained attention as time on task grows, is one of the most reproducible effects in cognitive psychology at group level. We deliberately do not turn that into an individual fatigue score: difference scores have poor test–retest reliability. What the long session gives is the chance to watch a person both fresh and tired inside the same task.

We will not overclaim the format. The systematic review of 34 studies by Ramos-Villagrasa and colleagues (2022) concludes that gamified assessment does not offer advantages sufficient to recommend it over conventional methods, with the exception of improved candidate reactions. Our honest formulation is therefore: comparable validity with substantially better candidate reactions. It is also worth knowing that a game format reduces but does not eliminate deliberate faking — shown in a controlled experiment with 171 participants.

And the current meta-analytic ordering does not support the story that longer and more immersive means more predictive. On the Sackett, Zhang, Berry and Lievens (2022) recalculation, structured interviews sit at .42, job knowledge tests at .40, work samples at .33, cognitive ability tests at .31, assessment centres at .29, situational judgement tests at .26 and conscientiousness at .19. High-fidelity simulations sit below the structured interview, not above it. We publish this table on our own site precisely because it does not flatter us.

Two design decisions are ours to defend rather than to benchmark. First, de-identification is architectural: the client receives codes and maps them to people itself, so we never receive a name, a gender or an age — a model cannot learn from protected attributes it has never seen. Second, the output is a fit assessment and a ranking, not a hire/don't-hire verdict; our own policy prohibits verdicts. We do not present that as a regulatory exemption, and nobody should read it as one: Annex III, point 4(a) of the AI Act covers systems used to evaluate candidates as such, and New York's definition turns on a simplified output — a score included — being relied on to substantially assist a hiring decision. Whether either regime applies to your use of our output is a question for your counsel, not a claim we make here.

Where the benchmark comes from

The short answer, and the substantive difference between us: our benchmark is built externally, from an occupational register and a lifecycle model, rather than from a sample of a client's own successful employees. Everything else follows from that choice — including its costs.

Concretely: 287 professions taken from the Russian Ministry of Labour occupational register and professional standards, crossed with 8 stages of the Adizes corporate lifecycle, giving 2,296 benchmark records across 34 functional areas. Each of those 2,296 cells carries a human-readable rationale — 2,296 distinct texts, 1.07 million characters, an average of 464 characters per cell, and 32% of them name the rule that fired. In a market where the EU AI Act, New York's rules and the CJEU's Dun & Bradstreet ruling all converge on explanation that a person can contest, a written reason attached to every cell is not decoration.

The lifecycle axis is not ornamental either. Across the 2,296 records, for 99% of the 287 professions the requirement ranges at Infancy and at Bureaucracy overlap by less than half, with a median range overlap of 0.40. Of the 153 outright non-intersections between the extreme stages, 152 fall on risk appetite. The span of the range centre across the lifecycle is 18.8 percentile points for risk, 15.4 for openness, 11.5 for conscientiousness and 9.0 for cognition; result focus barely moves. In plain terms: the same job title in a young company and in a mature one is, by requirement, a different job.

The benchmark also allows a shape that a total-match percentage cannot express. In 303 of the 2,296 cells — 13% — the benchmark permits a high risk appetite together with a low floor on cognition or progress monitoring. That is the method saying "fits on every parameter, and still warrants attention". A single summed score cannot represent that, in principle.

The practical consequence, and it cuts both ways. Because the benchmark is external, it works from day one and does not require you to have dozens of current employees in a role to calibrate against — and structurally it does not encode your existing staff composition into the target. The cost is equally real: an external register-based benchmark is a general model of a role, not a model of your company, and where your role genuinely differs from the register, the benchmark will be wrong in a way that a model trained on your own high performers would not be. That is a trade-off to make deliberately, not a feature.

How this page was made

This comparison is published by NeuroFrame — that is, by an interested party. That is exactly why every fact about the other product carries a link to a primary source, and why we assign no grades at all. We do not use evaluative adjectives about HireVue, and if you find one, tell us — we will remove it. There is no sentence in which an advantage of ours is derived from a shortcoming of theirs.

We work to three rules. Anything we say about another company's product traces to a primary source in the list at the foot of this page, with the date we opened it; a statement we cannot trace that way does not go on the page. Where we could not find something, we write that we did not find it on the pages we checked on 4 August 2026 — never that it does not exist, because vendors hold technical manuals and client documentation that are not public. And any complaint or litigation is reported strictly as the fact of a filing, with no inference about fault.

Our own numbers come from a single internal source of truth and nowhere else: comparison sample 14,850 people, predictive-validation sample 3,000+, CFI 0.96, Cronbach's α 0.69–0.77, test–retest above 0.83, session 30–60 minutes, library of 287 professions across 8 lifecycle stages. Where our own documentation disagrees with itself, we say so rather than pick the flattering figure.

External validity coefficients throughout this page use the Sackett, Zhang, Berry and Lievens (2022) recalculation, not Schmidt and Hunter (1998). The older values are still widely quoted — GMA at .51, work samples at .54 — but they carry a systematic over-correction for range restriction, and publishing them as current figures is simply an error.

If you find a factual mistake here, or a source that has moved, tell us and we will correct it with the correction visible. A page that quietly edits itself is worth less than one that shows its revisions.

Questions

In one paragraph: what actually differs between these two?

The unit of observation and where the benchmark comes from. HireVue is a platform where the game module sits alongside video interviews, simulations and technical checks; vendor materials describe cognitive packages of 7, 11 and 14 minutes and roughly 3 minutes per game. NeuroFrame is one continuous 30–60 minute simulation producing eight parameters, and the role requirement comes from an external library of 287 professions across 8 corporate lifecycle stages rather than from a sample of your existing staff. These are different instruments for different jobs, not a better one and a worse one.

How much of the candidate's time does this take?

For HireVue, per the vendor's own publications: roughly 3 minutes per game; a session from 7 minutes, sometimes 15–20; the candidate page advises setting aside 30 minutes, the product page says "each under 20 minutes", and the ebook whose content is dated 2020 lists packages of 7, 11 and 14 minutes. For NeuroFrame it is 30–60 minutes in one sitting. If you are screening tens of thousands of applications at the first step, those are different weight classes, and it is worth settling before a pilot rather than after.

Do you have a bias audit, the way HireVue does?

No. HireVue publishes DCI Consulting audit summaries under NYC Local Law 144 produced on 5 July 2023 and 5 July 2024, and the 2023 summary contains a separate section for the game-based instrument "Think – Shapedance, Numerosity". We have no formal bias audit and no adverse-impact-ratio report. It is also worth knowing how the New York rules allocate the duty: commissioning the annual audit and publishing its summary is the employer's or employment agency's obligation, and the rules do not put it on the assessment vendor. A vendor that commissions and publishes one is doing more than the rules ask of a vendor; whether it also covers your own obligation as an employer is a question for your counsel. If your procurement requires such a report in the file, that is an argument for HireVue.

We hire in the UAE. What about Arabic?

HireVue's regional presence is confirmed by the Emirates NBD partnership announced on 4 December 2024, but Arabic is not mentioned in that release, and on the vendor pages we checked on 4 August 2026 we did not find confirmation of Arabic localisation — worth asking the vendor directly. NeuroFrame has no Arabic at all: the assessment and report are in Russian and English, and the client library exists only in Russian. If an Arabic-speaking candidate flow matters, put the question to both of us.

Are there peer-reviewed publications behind either product?

The peer-reviewed paper we found for HireVue is Leutner, Codreanu, Brink and Bitsakis, Frontiers in Psychology, 18 January 2023 — convergent validity of the game-based cognitive assessment with ICAR at r = 0.5 (95% CI 0.43–0.56) and test–retest r = 0.68 on 102 participants; three of the four authors held a vendor affiliation, stated in the paper. NeuroFrame has no peer-reviewed publication of its own validation work — only internal figures. Worth noting separately: the authors of that 2023 paper themselves flag the absence of data linking game scores to job performance as a gap.

Can candidates prepare for an assessment like this?

That is a property of the format, not of a vendor. A discrete task with its own name and its own rule can in principle be found, dissected and rehearsed; a single continuous scenario changes the shape of that problem without removing it. About ourselves we will say it plainly: a game format reduces but does not remove deliberate faking — this was shown in a controlled experiment with 171 participants — and our continuous simulation is no exception to it.

Do you integrate with our ATS?

No, and this is not a roadmap gap — it follows from the architecture. We work with de-identified codes: the client receives codes and maps them to people, and we never receive a name, a gender or an age. There is nothing on our side to write into a candidate record. At a volume of thousands of candidates a week that is an operational problem, and HireVue, which publishes ATS integrations as part of its platform — the vendor's Q1 2025 product update names UKG, Greenhouse, Recruitee, Ashby and Team Tailor, plus a separate iCIMS Prime integration — is the more convenient choice in that scenario.