Understand the task types, compare the major vendors, learn what percentiles and trait profiles actually mean, and practice the formats without pretending there is one secret pass score.
By the Game Assessment Prep research teamUpdated July 14, 2026How we research
A game-based assessment is a hiring task whose interactions produce evidence about an ability, behavior, or preference. The candidate may compare quantities, remember a sequence, plan moves, withhold a response, recognize an emotion, or choose between uncertain rewards. The visual layer looks like a small game; the employer is interested in the pattern of decisions and performance underneath it.
“Gamified” is broader. It can describe a conventional test with progress bars and levels, or an immersive work simulation with game-like presentation. “Task-based” is another label: Arctic Shores now officially uses it for a product previously discussed as game-based. Cappfinity job simulations are often grouped into game guides, but they primarily combine scenarios, written work, and recorded responses rather than short cognitive games.
Employers use these formats to observe more than résumé history. That does not make every assessment valid, unbiased, or role-relevant by default. The quality depends on the task, comparison group, model, accommodations, and hiring decision built around it. Candidates should ask what a task measures and avoid both extremes: it is neither “just a game” nor a transparent mind-reading device.
A vendor name alone does not tell you which tasks an employer selected. Use the categories below to recognize the demand quickly, then practice the closest format rather than memorizing a single screen.
BART-style tasks ask whether to keep pumping for a larger reward or bank value before an uncertain pop. They can reflect risk adjustment and learning, not a universal preference for maximum caution. Practice pymetrics Balloons.
Sequence recall measures short-term span; n-back tasks require continuous updating. Aon gridChallenge also combines memory with executive attention. Try Digitspan, Flashback, or pymetrics Digits.
HireVue Numerosity, Aon digitChallenge, and Criteria Numbubbles put basic quantitative reasoning under a short clock. Arithmetic fluency and a stable scan method are trainable. Start with Numerosity.
Flanker, go/no-go, and rapid keypress tasks separate fast responding from controlled responding. False alarms matter as much as raw speed. Compare Arrows, Stop, and Keypresses.
Candidates label facial expressions, sometimes with contextual text. These tasks can be accuracy-scored, unlike a personality inventory. Practice Faces and E-Motions.
Tower, route, switch, and motion tasks examine forward planning, spatial reasoning, or deductive rules. Efficient solutions may matter alongside completion. Try Towers and Pathfinder.
Image choices, statement ratings, card decisions, and money exchanges describe preferences such as work style, risk learning, trust, or fairness. They do not share one “best” profile. Explore Portrait, Cards, and Money Exchange without trying to fake a personality.
Current ownership and product naming matter. This table separates official vendor facts from multi-source or single-source descriptions.
Last verified: July 2026
| Vendor / current name | Format | Typical duration | What it measures | Evidence behind this row | Practice |
|---|---|---|---|---|---|
| pymetrics (Harver) | 12-game behavioral battery | About 25 minutes | Behavioral and cognitive signals; roughly 90 measured behaviors is prep-industry consensus, not an official trait count |
| Practice all 12 |
| HireVue Games | Short cognitive, emotion, and work-style games, often bundled with recorded video | Usually 3–6 selected games over roughly 10–20 minutes | Numerical reasoning, memory, spatial processing, processing speed, emotion recognition, and work-style preferences |
| Practice HireVue formats |
| Arctic Shores | Task-Based Assessment (the company’s current name for the product) | Up to about 10 tasks; configuration varies | Dozens of traits across cognition and behavior |
| No direct simulator |
| Aon smartPredict | Four short challenges: switch, grid, digit, and motion | Roughly 5–9 minutes per challenge | Deductive logic, working memory, numeracy, executive attention, and planning |
| No direct simulator |
| Criteria Cognify | Three cognitive mini-games, with a six-game version available on request | About 10 minutes plus tutorials | Problem solving, numerical ability, and verbal ability; Emotify is the related emotional-intelligence product |
| No direct simulator |
| Cappfinity | Immersive strengths assessments and job simulations—not primarily a game-based vendor | Varies by employer simulation | Strengths, situational judgment, written and video responses, plus adaptive numerical testing where configured |
| No direct simulator |
Do not confuse these with every interactive assessment. McKinsey Solve is proprietary; Cappfinity centers on immersive job simulation; SHL offers interactive tests across a wider assessment catalog. If you are comparing paid prep providers, see our evidence-led comparisons of JobTestPrep alternatives, Graduates First alternatives, and 12minprep alternatives.
Cognitive tasks can measure working memory, processing speed, numerical reasoning, attention, inhibition, planning, or emotion recognition. Their outputs often include both accuracy and response time, so “faster” is not automatically better.
Behavioral tasks examine repeated choices: how risk changes after a loss, whether effort tracks reward, how someone allocates money, or which work-style statements they endorse. Those signals are descriptive. A cautious choice is not universally good, and a generous choice is not universally a pass.
Personality and preference games may have no objectively correct profile. Skill games still contain better and worse performance: a remembered sequence can be correct, a calculation can be wrong, and an impulsive press can be a false alarm.
The honest distinction is not “scored versus unscored vendor.” It is skill evidence versus preference evidence inside a battery. Our practice results show performance metrics for skill tasks and neutral trait descriptions for preference tasks.
A raw score records what happened: 18 correct answers, a nine-digit span, or a median response time. A percentile compares that result with a reference group. The 75th percentile means the result met or exceeded roughly 75% of that group; it does not mean 75% correct.
Norm groups matter. Change the comparison population and the percentile can change while the raw performance stays fixed. Our simulator percentiles use the Game Assessment Prep practice pool—not HireVue, Harver, or an employer’s applicant pool.
Several observations may be combined into a trait estimate: risk adjustment, effort allocation, attention, planning, or trust. An employer can then compare a combined profile with a role model. The exact production formulas, weights, and employer thresholds are generally not public.
That is why the same result can be interpreted differently by role. It is also why an insight report is not a hiring decision. A descriptive profile can be portable while the role comparison changes.
No evidence supports one 70% or 80% rule across game vendors, employers, and roles. Some processes use a cognitive cut score; others rank candidates, compare trait profiles, or combine the assessment with the application and interview. Anyone promising one universal number is hiding that variation.
Use practice scores to diagnose errors and improvement. Do not treat them as an offer probability. Our rebuilt guide to HireVue game scores explains the distinction in detail.
A retake interval is not shared across vendors. pymetrics officially allows replay about once every 330 days and can re-score an existing profile for another employer inside that window. McKinsey Solve is a separate proprietary assessment with a roughly 12-month reapplication interval consistently reported.
HireVue and employer-specific game retake rules vary by invitation. If no policy is published, do not invent one. Treat the live session carefully and ask support when a technical problem occurs. Our plans explain which practice feedback is available here.
Employer lists age badly. These examples preserve the confidence and transition language from our employer-specific research instead of repeating a 2019 customer list as current fact.
| Employer | 2026 evidence | Confidence |
|---|---|---|
| BCG | Archived US campus pages documented 12 pymetrics games, about 25 minutes, and a 48-hour window. Current US campus pages instead name a Consultant Career Assessment and online case without naming pymetrics. Experienced-hire candidates still report pymetrics, so the 2026 process is office- and channel-dependent. | OfficialMultiple sources |
| JPMorgan | Candidates consistently report a games stage before a separate HireVue video interview in many early-career pipelines. JPMorgan does not officially name the game vendor, so the pymetrics identification is triangulated from candidate reports rather than confirmed policy. | Multiple sources |
| Blackstone | Candidates consistently report the 12-game pymetrics battery early in analyst and associate processes, before recorded video. The employer’s exact weighting is not public. | Multiple sources |
| Unilever | The 2016–2019 pymetrics case study is historical. Unilever’s 2026 Future Leaders pages name no vendor, still describe a cognitive/emotional/social stage, and state that AI is not used as a screening tool. Treat old pymetrics branding as history, not current global policy. | Official |
| Bain | Bain’s own assessment guidance names Sova and TestGorilla, not pymetrics. The pymetrics claim is tied to older German-office candidate reports and remains single-source at its clearest point. | OfficialSingle source |
| NATO | NATO officially confirms HireVue interactive business games for longlisted Internship Programme applicants and a game-based first stage for YPP, while the general-vacancy process lists no games. | Official |
Practice reliably removes one large source of error: unfamiliarity. You stop spending working memory on controls and start applying a method. Numerical fluency, n-back routines, visual scanning, planning, and response inhibition can also improve with focused repetition.
Stable preferences are different. Repeating a trust or personality task to engineer a “desirable” answer is not skill training and can create an inconsistent profile. Use those simulations to understand the interaction and reflect, not to chase a good/bad percentile.
Run one diagnostic pass, repeat weak skill formats while protecting accuracy, and stop before fatigue trains careless behavior. The article Are HireVue Games Hard? ranks the trainability of all 12 HireVue-style formats.
These are independent reconstructions for format practice. They are not employer invitations, and their feedback uses our own rules and practice pool rather than a vendor’s private production model.
Begin with the invitation. Identify the vendor, named assessment, deadline, supported device, expected duration, and any accommodation or support route. Do not assume that an employer uses every game in a vendor’s library. A NATO internship invite, a Deloitte video interview, and a pymetrics battery can all involve digital assessment while requiring completely different preparation.
Play one representative task from each category named in the invite. Read instructions out loud, learn the controls, and note whether the task rewards accuracy, speed, efficient planning, or a preference choice. Use a desktop and physical keyboard when the real format recommends them.
Follow the current instruction rather than a remembered prep screen. Protect accuracy before accelerating, especially when a task changes rules. For personality and judgment formats, answer consistently with your real work behavior instead of trying to infer a universally preferred profile.
Use the vendor or employer support channel immediately. State the time, task, device, and visible error without claiming a score was lost. A technical interruption policy is employer-specific; reopening, refreshing, or creating another account without instructions can make the situation harder to resolve.
If the process also includes recorded questions, treat that as a separate skill. Our HireVue assessment guide covers interview setup and answer structure; game practice alone does not prepare a motivation or competency example.
It is a hiring assessment in which the candidate completes short interactive tasks designed to produce cognitive, behavioral, or preference data. Some tasks have objectively correct answers; others describe patterns such as risk, effort, trust, or work style.
A game-based assessment uses the task itself as the measurement instrument. A gamified assessment adds game-like presentation—progress, levels, or visual interaction—to a test or simulation. Vendors use the terms inconsistently, and Arctic Shores now prefers “task-based assessment.”
An employer can decide not to advance you, but most vendors do not publish one universal pass mark. Employers may compare a cognitive score or trait profile with a role model. McKinsey Solve is a separate proprietary assessment commonly treated as pass/fail, with roughly a 12-month reapplication interval reported across sources.
Yes for format familiarity and trainable skills such as mental arithmetic, working-memory routines, visual scanning, planning, and response control. Practice should not be used to manufacture a fake personality profile. Its largest benefit is removing avoidable errors caused by unfamiliar rules and controls.
Skill tasks can produce accuracy, speed, span, learning, or efficiency measures that may be standardized against a norm group. Behavioral tasks produce trait estimates or preference patterns. An employer may then compare several measures with a role-specific profile. The production formulas and hiring thresholds are usually not public.
The range depends on the provider and employer configuration. Criteria officially says Cognify takes about 10 minutes plus tutorials; short HireVue sets are consistently reported around 10–20 minutes; the 12-game pymetrics battery is about 25 minutes; larger task batteries or simulations can approach 45 minutes.
Confirmed and reported use varies by role, office, and cycle. NATO officially confirms games for NIP and YPP. Historical or candidate-supported examples include BCG, JPMorgan, Blackstone, Unilever, and P&G, but current vendor claims must be checked individually; Bain officially names Sova and TestGorilla instead.
A game interface is not automatically accurate or fair. Validity depends on the task, scoring model, role relevance, norm group, accessibility, and how the employer monitors adverse impact. Vendors publish validation claims, but candidates rarely see the employer-specific model. Request an accommodation when needed and judge broad fairness claims cautiously.
Platforms can record timing and interaction data, but exact detection rules are not public and vary. Do not use outside help unless the invitation explicitly allows it. NATO, for example, officially prohibits AI tools and outside help throughout recruitment and says detected use causes disqualification.
There is no universal restart policy. Stop guessing and follow the support route in the invitation. Record the time and error, take a screenshot if permitted, and contact the employer or vendor promptly. Do not repeatedly reopen the assessment unless instructed, because attempts may be limited.
Game Assessment Prep is independent and is not affiliated with HireVue, Harver/pymetrics, Arctic Shores, Aon, Criteria, Cappfinity, or the employers discussed. Processes change by role, office, and cycle; follow your invitation. Vendor and employer claims were reviewed in July 2026 and labeled by evidence strength.