pymetrics Cards Game: The Four Decks and How It Works
Luca from Candidate Falcon
Editorial Team
The pymetrics Cards game, often searched as the pymetrics card game, asks you to draw cards one at a time from four face-down decks. Every card wins you some money, and some cards also take money away. It is one of the 12 core pymetrics games, which Harver now runs as its gamified assessments after buying pymetrics in August 2022. Prep sites usually report it as the eighth game.
Facts below are labelled official (pymetrics or Harver material), prep sites (commercial descriptions) or not public.
The Cards game at a glance
| What is known | Source | |
|---|---|---|
| Task | Draw from four face-down decks; every card wins, some also lose | Prep sites |
| Start balance | $2,000 | Prep sites |
| Draws | About 80 on most prep sites; 30 to 40 in one walkthrough | Prep sites |
| Screen | A balance, gains and losses bars, and each card's result | Prep-site simulation |
| Deck payoffs | Not published | Not public |
| Origin | pymetrics says its games come from peer-reviewed studies; its patents do not name this task | Official |
What you see and do
- Instructions. Read them in full; your version may differ from prep-site reports.
- Four decks and a balance (prep sites). Four face-down decks, a starting balance of $2,000 and bars for your total gains and losses.
- Pick a deck (prep sites). The card shows what you won and, sometimes, what you lost.
- Your balance updates (prep sites). It is the start plus everything you have won, minus everything you have lost.
- Repeat until the game ends. Most prep sites report 80 draws. Whether the decks' payoffs change during the game is not public.
The psychology behind it: the Iowa Gambling Task
Prep sites unanimously describe the Iowa Gambling Task, published by Bechara, Damasio, Damasio and Anderson in 1994. In the original study, participants started with a $2,000 loan of play money and drew 100 cards from four decks. Two decks paid more per card but lost money over time; two paid less per card and came out ahead. Healthy participants shifted toward the steadier decks as the losses showed up. Patients with damage to the prefrontal cortex kept choosing the high-paying decks, which the authors called insensitivity to future consequences.
The classic schedule, per ten cards:
| Decks | Win per card | Losses per 10 cards | Net per 10 cards |
|---|---|---|---|
| A and B | $100 | $1,250 (A: frequent, smaller; B: one large) | −$250 |
| C and D | $50 | $250 (C: frequent, small; D: one) | +$250 |
pymetrics' $2,000 start matches the classic task, but its payoffs are not published and may differ (our inference).
What it is thought to measure
According to prep sites, Cards looks at learning from feedback, risk and decision making: whether, and how fast, your choices shift as gains and losses build up. Decision making, learning and risk tolerance are among the nine trait categories Harver lists. pymetrics does not publish which measures it takes or how it weighs them.
Overall, pymetrics compares your play across all the games with people who do well in the role and gives the employer a fit score, not a pass mark (official). See pymetrics results and the trait report.
Is there a right answer?
pymetrics says there is "no right or wrong answer" in its games, and Cards is a clear case. Unlike Arrows or Digits, no single draw is correct: you choose under uncertainty. In the research version two decks come out ahead, so researchers call them advantageous. But pymetrics' payoffs are not public, and its model compares your whole pattern of play with people in the role, so there is no published best way to play. More in pymetrics games answers.
Timing and length
- The whole set (official): about 25 minutes for the 12 games, 1 to 3 minutes each, with no pausing mid-game.
- Cards (prep sites): about 80 draws. No public source we found reports a time limit per draw.
How to approach it
pymetrics' advice is to be yourself. In Cards, that means deciding as you naturally would, with full attention on what each card tells you.
- Read every card. Look at the loss as well as the win.
- Watch the balance. It sums up everything so far.
- Judge a deck over several cards. One big win or loss says little on its own.
- Do not arrive with a script. The payoffs are not public, so a plan copied from a website may not fit your version, and it would not show how you decide.
- Set up calmly. A quiet place, a computer or the app, and no interruptions: games cannot be paused.
Common mistakes
- Judging decks only by the size of each win.
- Treating one large loss as the whole story of a deck.
- Clicking through quickly without reading each result.
- Assuming the amounts match the research task or a practice version.
- Quitting midway to restart: pymetrics' terms ask you not to stop a game halfway through.
A worked example
Six draws we made up, using the classic research amounts (not pymetrics' own) and neutral deck names.
| Draw | Deck | Won | Lost | Balance |
|---|---|---|---|---|
| Start | $2,000 | |||
| 1 | Deck 1 | $100 | $0 | $2,100 |
| 2 | Deck 1 | $100 | $0 | $2,200 |
| 3 | Deck 2 | $100 | $1,250 | $1,050 |
| 4 | Deck 3 | $50 | $0 | $1,100 |
| 5 | Deck 4 | $50 | $50 | $1,100 |
| 6 | Deck 1 | $100 | $250 | $950 |
After six cards, deck 1 looks good and deck 2 looks terrible. Both impressions rest on very few cards. In the classic schedule, deck 2's single big loss and deck 1's frequent smaller ones cost the same over ten cards: a net −$250 each. In the research task, what matters is how your picks change as evidence like this builds up.
How Candidate Falcon's Cards practice works
Candidate Falcon's independent pymetrics Cards practice game lets you get used to the format. It is not the real game and does not use pymetrics' payoffs, which are not public.
- 80 draws from $2,000, untimed, using the classic 1994 schedule.
- Decks move every session, so which position holds which schedule cannot be memorised; decks never run out.
- Tap, click or keys 1 to 4 (or A to D), with a short pause after each card so you can read it.
- A review of your final balance, how your picks moved in blocks of 20 draws and how each deck was set up. No pick is called right or wrong.
Practice results are not pymetrics or Harver scores. The pymetrics games come with every plan that includes Harver; see pricing. The free pymetrics guide covers the whole battery, and our balloon game example covers the other risk game.
FAQs
How many cards do you draw in the pymetrics card game?
pymetrics does not publish the number. Most prep sites report 80 draws and one walkthrough reports 30 to 40. The classic research task uses 100.
Which deck is best in the pymetrics Cards game?
Nobody outside pymetrics knows its real payoffs. In the classic Iowa Gambling Task, the two decks with smaller wins come out ahead over time, but pymetrics says there is no right or wrong way to play its games.
Is the pymetrics Cards game the Iowa Gambling Task?
Prep sites describe it that way, and the screen and $2,000 start match the classic task. pymetrics' patents do not name it, and its payoffs are not public.
Is the pymetrics Cards game timed?
pymetrics says each game takes 1 to 3 minutes and cannot be paused. No public source we found reports a time limit on each draw.
Sources
- Wilson et al. (2021), Building and Auditing Fair Algorithms, with pymetrics co-authors: games from peer-reviewed studies, the fit model.
- pymetrics patent WO2015200880A1: the patent family's task list.
- pymetrics candidate FAQ, employer-hosted copy (2018): timing, no pausing, no right or wrong answer.
- pymetrics End User Agreement (read October 2026): no stopping a game halfway through.
- Harver: gamified assessments and Harver acquires pymetrics (read October 2026).
- Prep-site descriptions (GraduatesFirst, MBB Prep, JobTestPrep): decks, $2,000 start, draw counts, screen.
- Bechara, A., Damasio, A. R., Damasio, H., and Anderson, S. W. (1994). Insensitivity to future consequences following damage to human prefrontal cortex. Cognition, 50(1–3), 7–15.
Candidate Falcon is independent and not affiliated with pymetrics, Harver or any employer.
