How these games work

What platforms look at, what we don't know and how to read the summary.

What these games are

These are six original games, built for this site, inspired by public experimental-psychology tasks: impulse control, working memory, planning, risk and reinforcement learning. They are not affiliated with any hiring platform and do not reproduce their games, parameters or reports.

Each game measures something specific to this one session (how many times you tapped, how long you took, how many mistakes you made), not a stable trait or a general ability. The summary describes what you did, not who you are.

What platforms tend to look at

Hiring platforms that use games like these usually train job-specific models on current employees: what counts as "good" changes per employer, and how that model was validated is rarely published.

The output is often a fit tier computed from a percentile, and a candidate can be screened out on that basis. The public bias audits that exist — including a 2025 audit of a game-based assessment platform, cited below — cover sex and race/ethnicity; almost never age or disability, even though reaction time changes with age and speed-, vision- or dexterity-based games can screen out people with a disability, who can request an alternative format.

In a large sample of applicants to high-stakes assessments, cognitive-ability test scores were lower on phones than on desktop; practicing beforehand helps a little, it guarantees nothing. Platforms such as pymetrics (Harver), Arctic Shores or HireVue use similar games in real hiring processes: this site is not affiliated with any of them and does not reproduce their games, parameters or reports.

What we don't know

We have no norms and no data on which model any given employer uses: we cannot tell you whether your result moves you toward or away from what a particular company is looking for, because that depends on a model we don't know.

One session is noisy: the repeatability of tasks like these, measured from one sitting to the next, tends to be modest. Tiredness, mood, background noise and prior practice change the result more than it might seem.

How to read the summary

Each bar shows a tendency of this one session between two neutral poles; it is not a grade or a percentile. The midpoint isn't "normal": depending on the game, the midpoint can be the mathematical optimum (Pump and Bank) or the balance between acting and waiting when unsure (Traffic Light, where the center corresponds to a decision criterion c = 0, with no bias toward tapping or withholding); in Four Decks, though, chance sits not at the midpoint but at the 0.75 reference mark.

Each bar's anchors are documented in the game itself: in Pump and Bank, the position is adjusted pumps divided by 128 (64 is the mathematical optimum with these balloons, not a norm); in Four Decks, it's the deck-switch rate, with a reference mark at 0.75 (switching at random); in Middle or Sides it's the median reaction time divided by 1500 ms; in Pegs and Discs it's the median latency before the first move divided by 15,000 ms; in Traffic Light the position combines how often you tapped the octagon and how many shapes you let pass, and the center corresponds to a decision criterion c = 0 (no bias toward or against tapping). Digit Series has no bar: it shows directly the longest series of digits you repeated.

No game measures your intelligence, personality, traits or ability; none predicts whether you will pass a hiring process, and none compares your result with other people's. Repeating a game a few times usually changes the result, so one session says nothing stable about you.

Device, age and accessibility

Tapping the screen adds roughly 60-70 ms to reaction time compared with a keyboard or mouse, and power-saving mode can lower the screen's refresh rate and make timings less precise: the summary flags it when that happens.

You can play with extra time (1.5× or 2×) in each game separately — the full circuit uses fixed timings, because reproducing that pace is part of what it practices — and turn on "Reduce motion and flashes" if motion bothers you or you have photosensitive epilepsy.

If a disability or condition affects your speed, vision or attention, you can ask the employer assessing you for timing, format or color-palette adjustments before you start: asking ahead of time is usually simpler than asking during the test.

Privacy

Your sessions are saved only in this browser for 24 hours (up to 30 days if you have the full report active): there is no account, your games are not uploaded to any server, and you can delete them at any time from the summary or the catalog. If you play inside a recruiter's evaluation, your play style (not your answers or timings) is sent to that company. Analytics events (game started, finished, abandoned or the circuit completed) are only sent if you accepted cookies, and never include scores, response times or answers.

Settings and the read notice are stored separately from your sessions, so deleting your games doesn't make you read the notice again.

Sources and attribution

These games are inspired by the following tasks and guidelines, without copying their exact content or algorithms:

• Lejuez et al. (2002) — the Balloon Analogue Risk Task
• Bechara et al. (1994) — the Iowa Gambling Task
• Eriksen & Eriksen (1974) — the flanker task
• Meiran (1996) — cued task switching
• Rogers & Monsell (1995) — task-switching costs
• Conway et al. (2005) — working-memory span tasks
• Shallice (1982) — the Tower of London task
• Robertson et al. (1997) — the Sustained Attention to Response Task
• Hedge et al. (2018) — reliability of cognitive-task measures
• Hausknecht et al. (2007) — practice and retest effects in testing
• Okabe & Ito (2008) — a color-blind-safe palette
• WCAG 2.2 (W3C) — timing, flashing, dragging and target-size criteria
• Arthur et al. (2014) — mobile vs. desktop scores in high-stakes assessments
• 2025 bias audit of a game-based assessment platform (PDF)