Pymetrics is a game-based assessment, owned by Harver since 2022, in which you play about a dozen short games (pumping balloons, remembering digits, sharing money with a partner, reading faces) instead of answering test questions. The software records how you play and compares your trait profile with people who do well in the role. Harver says it takes about 25 minutes; prep sites say 25 to 35. This guide explains each game, how the matching works, the retake rules and how to prepare, with original practice exercises.
Key facts
- Vendor
- Pymetrics (founded 2013), part of Harver since August 2022
- Games
- 12 core games; some roles add numerical or memory games
- Time
- about 25 min; 25–35 min reported; a few minutes per game
- Measures
- nine trait families, e.g. attention, effort, fairness, risk tolerance
- Answer format
- clicks, key presses, number entry, ratings; no knowledge questions
- Language
- mostly non-verbal
- Scoring
- matched to a model of successful employees in the role; no pass mark
- Retakes
- reported once every 330 days; results reused by other employers
Who uses pymetrics
Pymetrics began in 2013 as a "talent matching" platform built on cognitive-science experiments. Since Harver bought it in August 2022, it is sold as Harver's gamified soft-skills assessment, but invitations still show the pymetrics name and game designs.
The employer list changes from year to year. Prep sites report users among banks, consultancies, professional services firms and consumer-goods companies, mostly for graduate and early-career roles, and some employers have moved on to other game-based or case formats. The reliable signal is your invitation: it will say "pymetrics" or "Harver".
How pymetrics scoring works
Pymetrics is not graded like an aptitude test. Based on press coverage of an independent audit and prep sources:
- The employer builds a benchmark. At least 50 people who are successful in the target role play the games (MIT Technology Review, 2021).
- A model finds what is distinctive about them compared with a large reference sample of other players.
- The model is checked for adverse impact against the US "four-fifths rule" and rebuilt if pass rates differ too much between demographic groups. A 2021 audit by Northeastern University researchers (paid by pymetrics) found the code did this correctly; critics noted it did not show that the games predict job performance.
- You are compared with the model. One prep site says the employer sees a three-level recommendation (highly recommended, recommended, not recommended); others mention only a role-specific threshold. Pymetrics does not publish this, so treat the labels as reported.
What follows for you:
- There is no universally "good" profile. A cautious balloon player fits some roles, a bold one fits others.
- Noise hurts everywhere. Random clicking, misread rules, a lagging laptop or a distraction mid-game produce a profile that matches nobody.
Harver says candidates get results and feedback straight after finishing; prep sources describe a trait report, not a score or a pass/fail notice.
The 12 pymetrics games
Mechanics combine several prep sources; disagreements are noted. Most games derive from laboratory tasks (last column), which tells you what is being measured.
| Game | What you do | Traits (reported) | Lab task it resembles |
|---|---|---|---|
| Balloon | Pump for money per pump; bank whenever you like; a burst loses that balloon's money. Colours burst differently. One source says 39 balloons | Risk tolerance | Balloon Analogue Risk Task |
| Keypress | Tap one key as fast as you can for a few seconds, between start and stop cues | Effort, focus | Finger tapping |
| Digits | Type back digits shown one at a time; sequences get longer | Attention, memory | Digit span |
| Arrows | Answer the direction of the middle or the side arrows, depending on colour. Sources disagree on the colours; one says about 135 rounds | Attention, learning | Flanker with task switching |
| Lengths | Decide quickly whether a cartoon face has the long or the short mouth; the difference is tiny | Attention, learning | Probabilistic reward task |
| Cards | Draw from four decks with hidden reward and penalty patterns; one source says you start with $2,000 | Learning, risk | Iowa Gambling Task |
| Tower | Recreate a target arrangement of coloured discs on three pegs in as few moves as possible; time before the first move is recorded | Planning, decision making | Tower of London |
| Money Exchange 1 | Send part of $10 to a partner; it is tripled; the partner returns some; you rate fairness | Fairness, trust | Trust game |
| Money Exchange 2 | You and a partner get $5 each and one gets an extra $5; share or take, then rate fairness | Generosity, fairness | Dictator game |
| Hard or Easy | Pick an easy task for a small reward or a hard one (many more key presses) for a bigger reward, with a stated chance that the round pays | Effort, risk | Effort Expenditure for Rewards Task |
| Stop | Press only when a specified colour or shape appears | Attention, impulse control | Go/no-go |
| Faces | Name the emotion in a face, sometimes with a short caption | Emotion | Emotion recognition |
Extra games. Several sources say some employers add about four numerical or memory games but name them differently: one lists Magnitudes (comparing dots and fractions), Sequences (missing number), Shapes (find a small shape in a complex pattern) and an N-back letter game; another groups four extras under "numerical dexterity" and "quantitative reasoning". Harver also sells separate gamified cognitive tests that may come in the same invitation. These have right answers, so ordinary numerical and inductive practice helps.
For how pymetrics compares with Arctic Shores, Aon and McKinsey Solve, see the game-based assessment overview. It already works through Balloon expected value, a two-deck Cards example and an Arrows colour drill, so the exercises below cover other games.
Original practice exercises
All exercises are original; amounts, timings and rules are invented to practise the mechanics, not pymetrics' actual parameters.
Balloon: learning each colour
So far, teal balloons burst at 5, 7 and 9 pumps and one teal was banked safely at 4. Amber balloons burst at 24 and 31 and one was banked at 18. What is a sensible plan for the next teal and amber balloons?
Answer: stop teal at about 4–5 pumps and amber at about 18–22. - Teal's earliest burst was 5, and 4 has been safe. Going to 6 would have lost one of three teal balloons so far. - Amber has never burst below 24; stopping a little under that keeps most of the upside. - The lesson is to treat each colour separately and update as bursts arrive. How boldly you then play is your style, and that is what is recorded.
Digits: chunking
Read each line once at about one digit per second, look away, write it down: (1) `7 2 9 4 1` (2) `3 8 0 6 5 2 7` (3) `9 1 4 4 8 6 3 0`.
Answer and method: check against the lines. Most people hold five to seven single digits; beyond that, chunk: "380, 65, 27" or "91, 44, 86, 30". Repeated digits are anchors. Rehearse the chunks silently while the next digit appears. Ten lines a day for three days usually adds a digit or two to your span.
Tower: minimum moves
Rules for these items: three pegs, only the top disc moves, no height limit, discs listed bottom to top.
Item 1. Start: peg 1 = Red, Green; peg 2 = Blue; peg 3 empty. Target: peg 3 = Blue, Red, Green. Minimum moves?
Answer: 4. Blue 2 → 3; Green 1 → 2; Red 1 → 3; Green 2 → 3. - Every disc must move once, and Green must move twice: Red, beneath it, has to reach peg 3 first.
Item 2. Start: peg 1 = Blue, Red, Green; pegs 2 and 3 empty. Target: peg 1 = Green, Red, Blue (reversed). Minimum moves?
Answer: 7. Green → 2; Red → 3; Blue → 3 (on Red); Green → 1; Blue → 2; Red → 1; Blue → 1. - Lower bound: all three discs must leave peg 1 and return, so at least 6. Six fails because, with three discs on two spare pegs, Blue (out last) must sit on Green or Red and has to move once more before Red can go back. - Lesson: plan the full sequence before the first click. Tower records both move count and thinking time, so a short pause is normal; changing plans mid-way adds moves.
Hard or Easy: reward per second
Invented version: the easy task takes 7 seconds and pays $1.00; the hard task takes 21 seconds. A banner shows the chance that the round pays, and it applies to whichever task you pick.
| Round | Hard task pays | Chance the round pays |
|---|---|---|
| A | $2.00 | 88 % |
| B | $3.50 | 12 % |
| C | $4.20 | 50 % |
In which rounds does the hard task earn more per second?
Answer: B and C. - Easy: $1.00 ÷ 7 ≈ $0.143 per second. Hard: A ≈ $0.095, B ≈ $0.167, C = $0.200. - The payout chance multiplies both options equally, so it does not change which is better per second; it changes how much is at stake, which is where effort and risk preferences show. There is no correct policy; the exercise makes the trade-off visible before you meet it under time pressure.
Stop: holding back
Rule: press only for a green circle. Sequence: green circle, green circle, green circle, red circle, green circle, green square, green circle, blue circle. Which items get a press?
Answer: 1, 2, 3, 5 and 7. Item 6 is green but a square. A run of presses builds the urge to press, and a near-miss (right colour, wrong shape) slips through. Check both features every time; a slightly slower press beats a false alarm.
Faces: expression and caption
- Inner eyebrow ends raised and drawn together, lip corners pulled down. No caption.
- Upper lids raised, eyebrows high and curved, jaw dropped, no smile. Caption: "Her name is called out as the winner."
- Eyebrows lowered and pulled together, lips pressed tight. Caption: "Someone has just pushed in front of him."
Answer: 1 sadness; 2 surprise; 3 anger. In item 2 the caption hints at joy, but the face shows surprise. Read the face first and use the caption to decide between close options; prep sources differ on how captions are weighed. The emotion recognition guide lists the cues for each emotion.
Traps and how to avoid them
- Skipping instruction screens. Each game has one rule you cannot infer (which colour means "side arrows", which key to press).
- One pump count for every balloon. Colours behave differently.
- Chasing a "right" profile. Acting generous in Money Exchange and ruthless in Hard or Easy creates an inconsistent pattern across games that measure related traits.
- Losing focus on repetitive games. Lengths, Stop and Arrows reward steady attention over many rounds.
- Weak hardware. Trackpads and laggy phones add errors on speed games.
How to prepare for pymetrics
Harver's position is that the behavioural games need no study. That holds for style traits, but knowing the mechanics removes noise, and the extra numerical games are trainable.
| Day | What to do | Time |
|---|---|---|
| 1 | Read the game table; do the Tower, Hard or Easy and Stop exercises; read the game-based overview for Balloon, Cards and Arrows | 45 min |
| 2 | Ten digit lines with chunking; emotion cue review; test browser and keyboard. If numerical games are listed, add number series and quick mental arithmetic | 45 min |
On the day: - Laptop with keyboard, a current Chrome, Firefox, Safari or Edge, stable connection. - Block 40 minutes. Prep sources say you can pause between games but not within one, and that deadlines can be short (one mentions 48 hours). - Raise any accommodation need (colour blindness, ADD/ADHD, dyslexia) with the employer before the first game. - Play attentively and as yourself.
Sources
- Harver, "Harver Acquires pymetrics" (2022): https://harver.com/?p=36917; deal date August 2022 per PitchBook
- HCM Technology Report on the acquisition (founded 2013)
- pymetrics.com candidates page (redirect notice)
- Harver, gamified assessments (about 25 minutes, "12+" games, nine traits, non-verbal, immediate feedback)
- MIT Technology Review (2021), audit coverage (50 employees, 12 games, four-fifths rule, criticisms)
- Wilson, C., et al. (2021). Building and auditing fair algorithms: A case study in candidate screening. FAccT '21, 666–677: https://www.ccs.neu.edu/home/amislove/publications/Pymetrics-FAccT.pdf
- Psychometric Success (12 core + four numerical games, 25–35 minutes)
- GraduatesFirst (mechanics, $10 tripled, $2,000 in Cards, 330-day rule, pausing, browsers, PC and smartphone support)
- MyConsultingCoach, pymetrics guide (~330-day sharing, laptop advice)
- RoadToOffer, pymetrics test guide (39 balloons, ~135 Arrows rounds, three-level recommendation, 48-hour deadline, extra modules)
- PrepLounge (four extra numerical games)
- Lab tasks: Lejuez et al. (2002), BART, J. Exp. Psych.: Applied 8; Bechara et al. (1994), Iowa Gambling Task, Cognition 50; Berg, Dickhaut & McCabe (1995), trust game, Games and Economic Behavior 10; Treadway et al. (2009), EEfRT, PLoS ONE 4(8); Pizzagalli et al. (2005), probabilistic reward task, Biological Psychiatry 57; Shallice (1982), Tower of London, Phil. Trans. R. Soc. B 298; Eriksen & Eriksen (1974), flanker task, Perception & Psychophysics 16.
MSc Psychology, 7 years of test taking.
Last updated: Oct 1, 2026