Pymetrics-style game practice

Each game takes 1 to 3 minutes, cannot be paused, and breaks are allowed between games; "there is no right or wrong answer"

Source · 2018 candidate FAQ reproduced in an employer guide.

Five arrows flash and the arrow colour sets the rule: report the centre arrow or the side arrows

Source · We use a frame style and label instead of colour.

About 135 trials, about 3 minutes

Source · Prep sites.

Patent feature "All Switching, Accuracy"

Source

Flanker task: congruent versus incongruent side arrows; cost in reaction time and errors

Source · Eriksen and Eriksen 1974, Perception and Psychophysics 16:143.

Cued rule switching: switch cost in reaction time and errors

Source · Rogers and Monsell 1995, JEP: General 124:207.

The cue, arrow and response timings and a 50 percent switch rate are Kvanta choices; the vendor's values are not public

Not a vendor fact. Chosen to match the academic protocol and the vendor's 1 to 3 minute guidance.

A balloon game in the 12-game set: pump a balloon for money, bank at any time, a pop loses that balloon's money

Source · Harver lists risk tolerance among its trait categories; the patent names the number of clicks as a feature.

The balloon game appears in the vendor's patent as a risk-taking and impulsivity measure, with number of clicks as a feature

Source · US9842314.

About 39 balloons

Source · Candidate and prep-site reports; may differ by employer. Some report colours with different burst risk.

Each balloon bursts at a hidden pump number drawn uniformly from 1 to 128

Source · Lejuez et al. 2002, J Exp Psychol Appl 8:75.

Standard metric: adjusted average pumps, the mean pumps on balloons that did not burst; also the number of bursts

Source · Retest correlation about .77 across days (White, Lejuez and de Wit 2008), so practice barely moves it.

Retest correlation about .77 with no mean change across days

Source

The value of a pump (5p) and the number of practice balloons (2) are Kvanta choices; the vendor's values are not public

Not a vendor fact. Chosen to match the academic protocol and the vendor's 1 to 3 minute guidance.

Draw from four decks; each card wins or loses money

Source · Not found in the patents reviewed.

About 30 to 40 draws in the vendor-style version

Source · Used in the battery version (40 draws).

Iowa Gambling Task: 100 draws; decks A and B net minus 250 per 10 cards; decks C and D net plus 250; net score per block of 20

Source · Bechara et al. 1994, Cognition 50:7.

Practice effects are strong because players learn the deck rules; the standard score has retest r of about .37; randomise deck schedules and positions each session

Source · We re-randomise deck positions and the loss positions every session.

The starting pretend money (2,000) and the card amounts are Kvanta choices; the vendor's values are not public

Not a vendor fact. Chosen to match the academic protocol and the vendor's 1 to 3 minute guidance.

Digits flash and are typed back; the list grows by one after each success

Source · Measures working memory.

Forward digit span: length plus one after a success; metric is the longest list recalled

Source · Wechsler digit span.

The start length (3), one digit per second and two misses ending the game are Kvanta choices; the vendor's values are not public

Not a vendor fact. Chosen to match the academic protocol and the vendor's 1 to 3 minute guidance.

Hard or easy: each round choose an easy task (few presses, small reward) or a hard one (many presses, bigger reward) with a stated win probability

Source · Measures effort and motivation.

Easy task 30 presses in 7 seconds for 1 pound; hard task 100 presses in 21 seconds for 1.24 to 4.30 pounds; win probability 12, 50 or 88 percent

Source · Treadway et al. 2009, PLoS ONE 4:e6598. Original amounts were dollars.

Metric: share of hard choices by probability and reward

Source · Original analysis used GEE regression; we report shares and slopes.

The vendor's own task parameters (presses, durations, rewards, odds) are not public

Source · The patent names slope and inflection-point features only.

The 8-minute time budget (3 minutes and half-length tasks in the battery version) are Kvanta choices; the vendor's values are not public

Not a vendor fact. Chosen to match the academic protocol and the vendor's 1 to 3 minute guidance.

A face is shown, sometimes with a short story; pick the emotion from 4 options; the patent credits identifying the emotion without the story

Source · We show no story by default.

Metric: accuracy and the unbiased hit rate

Source · Wagner 1993, Journal of Nonverbal Behavior 17:3.

Faces are stylised parametric drawings, not photographs: weaker in fidelity than photo sets, but no licences, no consent issues and unlimited identities

A documented Kvanta trade-off, not a vendor fact. Academic face sets often restrict commercial use.

The number of faces (36) and the seven emotions are Kvanta choices; the vendor's values are not public

Not a vendor fact. Chosen to match the academic protocol and the vendor's 1 to 3 minute guidance.

A cartoon face flashes; say whether the mouth is short or long; money for some correct answers only; patent "Reward Learning Task" with short or long mouths and arrow keys

Source

Three blocks of 100 trials; mouths of 11.5 and 13 mm shown for 100 ms; one mouth rewarded three times as often

Source · Pizzagalli et al. 2005, Biological Psychiatry 57:319.

Metrics: response bias log b and discriminability log d

Source

Learning-based games need a re-randomised setup each session; here which mouth is rewarded more often changes every session

Source · Practice effects are strong when the schedule is fixed.

The reward size (20p), 30 and 10 rewards per block, and the trial timings are Kvanta choices; the vendor's values are not public

Not a vendor fact. Chosen to match the academic protocol and the vendor's 1 to 3 minute guidance.

The battery version uses one block of 100 trials to fit the 1 to 3 minute guidance

Source · Shortened from the academic three blocks.

A "money now or later" game was reported by JPMorgan candidates

Source · Press report (efinancialcareers, 2019/2020).

The patent describes a Future Discounting task that yields a discount rate

Source

Monetary choice questionnaire design: 27 choices, nine discount-rate levels by three reward sizes; metric is the hyperbolic discount rate k

Source · Kirby and Marakovic 1996, Psychonomic Bulletin and Review 3:100. Items are generated fresh each session from the published design, not copied.

The generated amounts and delays are Kvanta choices; the vendor's values are not public

Not a vendor fact. Chosen to match the academic protocol and the vendor's 1 to 3 minute guidance.

A second money game: choose a split between yourself and the partner, then rate fairness; patent feature "Amount 2"

Source

Dictator game: split money; the partner cannot respond; the metric is the amount given

Source · Forsythe et al. 1994, Games and Economic Behavior 6:347.

Some prep sites describe ultimatum-style elements; whether the vendor game is a dictator game is not confirmed

Source · We build the dictator game.

The endowment (10 pounds) and 3 rounds are Kvanta choices; the vendor's values are not public

Not a vendor fact. Chosen to match the academic protocol and the vendor's 1 to 3 minute guidance.

Press the space bar for one shape and withhold for another

Source · The vendor uses colours (red and green circles); we use shapes so it works in black and white and for colour-blind players.

Go/no-go: respond to the frequent target and withhold on the other; metrics are commission errors, omissions, go reaction time and d-prime

Source · Stanislaw and Todorov 1999, Behavior Research Methods 31:137.

The number of trials, the share of no-go trials (25 percent) and the stimulus timings are Kvanta choices; the vendor's values are not public

Not a vendor fact. Chosen to match the academic protocol and the vendor's 1 to 3 minute guidance.

Twelve core games, about 25 minutes in total

Source · Harver, the vendor's owner since 2022.

Players missing more than two games are dropped from scoring

Source · FAccT 2021 audit, co-written by the vendor's staff.

Output tiers: 70th percentile and above Highly Recommended, 50th to 70th Recommended, below 50th Not Recommended; "the games are not meant to be won or lost"

Source · Each employer has its own private model trained on its own staff. Kvanta shows the format only, against an example profile.

The game list (balloon, two money games, tapping, hard or easy, digits, stop, arrows, mouth lengths, cards, faces, towers) follows the patent and prep-site descriptions

Source · Plus candidate and prep-site reports (road to offer, Oxford University careers service).

Used at JPMorgan (reported since 2019)

Source · Press report. Standard Chartered and Blackstone are candidate-reported only.

Standard Chartered and Blackstone are reported to use the same vendor

Source · Candidate and prep-site reports. Banks change vendors often; check the current cycle.

Per-game counts are shortened from the academic protocols to fit the vendor's 1 to 3 minutes per game (for example 24 balloons, one block of 100 mouth trials, 40 card draws, a 150 second effort budget)

Source · Standalone game presets use the fuller academic protocols.

Faces are stylised drawings, not photographs, and can be skipped; skipped games count as missing

Kvanta trade-off and neurodiversity opt-out.

Tap a key as fast as possible; the patent version uses 60 seconds of tapping

Source · Measures effort and motor speed.

Start and stop cues alternate during tapping

Source · Candidate reports.

Metric: taps per interval

Source · Standard finger-tapping measure.

Six 10-second intervals with 3-second rests are Kvanta choices; the vendor's values are not public

Not a vendor fact. Chosen to match the academic protocol and the vendor's 1 to 3 minute guidance.