pymetrics Lengths Game Explained: Short or Long Mouths

In the pymetrics Lengths game, a simple cartoon face flashes on screen and you decide whether its mouth was short or long. The two lengths differ only slightly, and some correct answers win a little money while others show nothing. It is based on a research task about how rewards shape decisions, which is why a game that looks simple is hard to read.

pymetrics is owned by Harver, which completed the acquisition in August 2022 and now sells the games as its gamified assessments. This guide covers what you see, the research behind the game, what it is thought to measure and what to expect.

The short answer

A face appears with a short or a long mouth, shown only briefly.
You press the left arrow for short or the right arrow for long.
Some correct answers win money, and one length is rewarded more often than the other. You are not told which.
Nothing marks your answers right or wrong.

What you see and do

Official (pymetrics' patent): the patent calls it a reward learning task. Digital faces have a short or a long mouth, with a subtle difference between them. The right arrow means long and the left arrow means short. Correct choices earn money on an uneven schedule, so one length pays more often.
Prep sites report: a cartoon face flashes and you classify its mouth. Some correct answers show a small amount of money (one site reports $0.20); other correct answers and all wrong answers show nothing, and there are no right or wrong marks. Prep sites report either 60 or 90 faces.
Not known: the number of faces, the reward ratio and amount, how long the mouth is shown and the exact mouth sizes. Public sources say less about Lengths than about most other games. Some prep sites simulate it with their own pictures and keys.

The research task behind it

Lengths follows the probabilistic reward task of Pizzagalli, Jahn and O'Shea (2005). In the original:

Faces: cartoon faces with a short mouth (11.5 mm) or a long mouth (13 mm).
Timing: the mouth appeared for just 100 milliseconds after a half-second delay, and the face stayed until the participant answered.
Trials: 300 faces in three blocks of 100.
Uneven rewards: in each block, 30 correct answers to one length (the "rich" one) and 10 correct answers to the other (the "lean" one) were rewarded, a 3 to 1 ratio. Each reward was 5 cents.

Because the difference is so hard to see, people tend to start answering with the rich length when they are unsure. Researchers measure two things:

Response bias (log b): how far you lean towards the length that pays more often.
Discriminability (log d): how well you tell the two lengths apart, regardless of that lean.

The authors designed it to measure responsiveness to reward and used it to study anhedonia, a reduced response to reward. Leaning towards the rich length is normal, not a mistake.

What it is thought to measure

Harver lists learning, attention and decision making among the nine trait areas its games cover, without saying which game feeds which. In research terms, Lengths looks at:

How rewards shape your choices, through response bias.
Attention to fine detail, through discriminability.
Learning over time, by comparing early and late blocks.

pymetrics compares your game measures with those of people who do well in the job. It does not publish how it scores Lengths. For how results are reported back, see pymetrics results and the trait report.

Is there a right answer?

Partly. Each mouth really is short or long, so every face has a correct answer and accuracy can be measured. But the game's main interest, as in the research task, is the lean that rewards create, and pymetrics says there is no right or wrong answer in its games. There is no published target for accuracy or bias.

How long it takes

pymetrics says each game takes 1 to 3 minutes, and a game cannot be paused once it starts. With 60 to 90 faces reported, each one passes quickly, so you need to be looking at the right spot when the mouth appears.

A worked example

These numbers are for illustration only. Suppose long mouths are the rich length, although you do not know it, and you see 30 faces: 15 long and 15 short.

Of the 15 long faces, you call 13 long and 2 short.
Of the 15 short faces, you call 10 short and 5 long.
Some of your correct answers show money, more often after long faces than short ones.

Your accuracy is 23 out of 30, or 77%. You answered "long" 18 times although only 15 faces were long: a lean towards the rich length. Using the research formulas, your response bias is about 0.26 (above 0 means a lean towards the rich length) and your discriminability about 0.56 (0 would mean guessing).

That is neither "good" nor "bad": the rewards pulled your answers a little, and you saw the difference most of the time.

Tips

Fix your eyes on the lower middle of the face, where the mouth will appear, before it shows.
Answer on your first impression. Staring at the face after the mouth has gone will not bring it back.
Expect to be unsure on many faces. The difference is meant to be hard to see.
Do not read silence as a wrong answer. Most correct answers show nothing either.

Common mistakes

Counting rewards to find a pattern instead of watching the faces.
Looking down at the keyboard when the mouth appears.
Answering the same length for a long stretch whenever you are unsure.
Confusing it with the Faces game, which is about recognising emotions.

Practising the format

Candidate Falcon's Lengths practice is an independent version built from the patent, the research task and prep-site descriptions, with our own simple face. You see 90 faces in three blocks of 30, and the mouth shows for 0.1 seconds, as in the research task. You press the left arrow for short or the right arrow for long, or tap Short or Long. A rewarded answer shows +$0.20 and every other answer shows nothing. The rich length is picked at random each run and paid three times as often.

The review reveals which length paid more and shows your accuracy, response bias and discriminability, overall and block by block, in plain words. It is not a pymetrics score or trait report.

It is not pymetrics' game. Playing needs a plan that includes Harver, which comes with the pymetrics-style games; see pricing. The Cards game is the other pymetrics game about learning from rewards, and the free pymetrics guide covers all 12.

FAQs

What is the pymetrics lengths game?

It is a version of the probabilistic reward task: cartoon faces with a short or a long mouth, left arrow for short and right arrow for long, and correct answers paid on an uneven schedule.

Why do I only win money sometimes in the Lengths game?

Only some correct answers are rewarded, and one length more often than the other. Other answers show nothing, so you never see whether you were right. That is by design.

Is there a right answer in the pymetrics lengths game?

Each face is short or long, so accuracy can be measured. pymetrics still says its games have no right or wrong answers, and it publishes no target for accuracy or bias.

How many faces are in the Lengths game?

pymetrics does not publish a count. Prep sites report 60 or 90; the research task used 300.

Sources

pymetrics patent WO2015200880A1: the reward learning task with short and long mouths.
Wilson et al. (2021), Building and Auditing Fair Algorithms, with pymetrics co-authors: how game measures are compared with high performers.
Harver: gamified assessments (read October 2026): trait areas and the games after the acquisition.
pymetrics candidate FAQ, employer-hosted copy (2018): no right or wrong answer and timing.
Pizzagalli, D. A., Jahn, A. L. and O'Shea, J. P. (2005). Toward an objective characterization of an anhedonic phenotype: a signal-detection approach. Biological Psychiatry, 57(4), 319–327.
Prep sites (descriptions, not official): GraduatesFirst and Psychometric Success.

Candidate Falcon is independent and not affiliated with pymetrics, Harver or any employer.