What it is
Count is a numerosity estimation game in the Senso collection of perceptual precision games. On each trial a field of dots appears on screen for about 1.4 seconds, then disappears behind a visual mask. Your task is to type how many dots you think were there. For larger arrays the mask cuts you off before a careful tally is possible, so you lean on your intuitive sense of quantity.
The dot count ranges from 8 to 40 dots. What makes Count scientifically rigorous is how those dots are drawn: trials alternate between a constant-total-area layout and a constant-dot-size layout. This decorrelates the total inked area, the average dot density, and the number of dots from one another, so the only reliable cue to 'how many' is number itself. You cannot cheat by judging 'how much stuff is on screen' β you have to estimate numerosity directly.
How to play
Each session consists of five rounds. A fixation cue appears, then the dot field shows for about 1.4 seconds, then the mask replaces it. Type your estimate in the input box and confirm. You receive immediate feedback showing the actual count alongside your guess and the points earned for that round. After five rounds your total score out of 50 is displayed.
There is no time pressure once the mask appears β take a moment to anchor your impression before typing. Resist the urge to hedge towards round numbers; your first gut feeling tends to be more accurate than a second-guessed compromise. The difficulty is calibrated so that even practised estimators have room to improve.
The science
The ability to rapidly judge 'how many' without counting is called the Approximate Number Sense (ANS). It is one of the oldest cognitive faculties studied in numerical cognition research: infants a few days old can discriminate arrays that differ in number, and the same capacity has been documented in fish, birds, monkeys, and many other species. The ANS is a nonverbal, pre-symbolic system that operates in parallel with the language-based number system humans learn in childhood.
The ANS obeys Weber's law: the ease of distinguishing two quantities depends on their ratio, not their absolute difference. Telling apart 10 dots from 20 dots is roughly as easy as telling apart 40 dots from 80 dots, because both pairs share the same 1:2 ratio. Each individual has a characteristic Weber fraction β a measure of their number acuity β that describes how large a ratio difference they need before they reliably notice it. A lower Weber fraction means sharper numerosity perception. Count's scoring metric is built directly on this principle: your error is measured as the absolute log-base-2 ratio of your guess to the actual count, so an overestimate and an underestimate of the same ratio magnitude cost the same number of points.
The intraparietal sulcus (IPS), a region of posterior parietal cortex, is strongly and consistently implicated in numerosity processing across neuroimaging studies. Patients with damage to parietal regions often show selective impairments in approximate number judgements while retaining other cognitive abilities. The IPS appears to contain neurons tuned to preferred numerosities, analogous to orientation-selective neurons in visual cortex.
Count also touches on subitizing β the fast, exact, and virtually effortless apprehension of very small quantities (typically 1 to 4 items). When a display contains four or fewer dots, most people report their count with high confidence and near-zero error; reaction times are flat across that range. Above roughly four items, performance shifts to the noisier estimation regime governed by the ANS. Research has shown that ANS acuity measured by tasks like Count correlates with formal mathematical achievement, though the relationship is distinct from general intelligence and is thought to reflect a foundational perceptual substrate that symbolic math builds upon rather than replaces.
Scoring explained
Each round is worth up to 10 points. Your score for a round is calculated as 10 minus a penalty derived from your log-ratio error: error = |log2(guess / actual)|. You earn exactly half the round's points β 5 out of 10 β when your error reaches 0.17 log2 units, which corresponds to being off by about 12 percent. A perfect guess scores the full 10; an estimate more than roughly 25 percent off in either direction scores progressively less.
Over five rounds the maximum total is 50 points. Because the metric is symmetric in log space, overshooting and undershooting by the same factor are penalised equally. Tracking your score across sessions is a direct window onto your Weber fraction: consistent improvement means your ANS acuity is sharpening, either through genuine perceptual learning or through a reduction in response bias (such as the common tendency to underestimate large dot arrays).
Tips to improve
- Look at the whole field at once rather than scanning β your visual system integrates density and spatial extent in parallel, and saccading wastes the limited exposure time.
- Anchor to a sub-region first. If you briefly glimpse a cluster of roughly ten dots in one quarter of the screen, scale up based on how the rest of the field compares to that anchor.
- Notice your consistent bias. Most people underestimate large arrays and overestimate very sparse ones. If you know you tend to go low, nudge your response upward before submitting.
- Do not round to the nearest five or ten. The scoring function penalises ratio errors symmetrically, so precise estimates that happen to be odd numbers are no worse than round ones β but rounded guesses cluster around a narrower range and will sometimes miss by more.
- Commit to your first impression. The brief exposure and mask leave little time to tally larger arrays, and post-hoc rationalisation is more likely to pull you away from an accurate initial estimate than to improve it.
- Train consistently rather than in long isolated bursts. Perceptual learning in numerosity tasks accumulates across sessions; the neural populations in the intraparietal sulcus that encode numerosity benefit from spaced exposure just as motor skills do.
FAQ
Why does the dot size change between trials?
Dot size is varied deliberately to prevent you from using total inked area as a proxy for number. If all dots were the same size, a larger array would always look 'heavier' or 'denser', and you could judge quantity indirectly by judging visual mass. By alternating between constant-total-area layouts (where more dots means smaller dots) and constant-dot-size layouts (where more dots means a denser field), Count ensures that neither area nor density alone predicts number. You are forced to estimate numerosity itself.
What is the Approximate Number Sense and can it be trained?
The Approximate Number Sense is an evolutionarily ancient, nonverbal system for estimating quantity without counting. It is present from birth and is shared with many animal species. Research on whether the ANS can be improved through training is ongoing; some studies find genuine improvements in Weber fraction with practice, while others attribute gains to reduced response bias rather than sharper perception. Either outcome benefits your score, and reducing systematic bias is itself a form of cognitive calibration.
Why does the game use a log-ratio error metric instead of a simple difference?
Weber's law tells us that the subjective difficulty of a numerical judgment scales with the ratio of the quantities involved, not their absolute difference. Being off by 5 when the answer is 10 is a much larger perceptual error than being off by 5 when the answer is 80. The log-ratio metric β |log2(guess / actual)| β captures this ratio-based scale correctly, so your score reflects genuine perceptual precision rather than penalising large-number trials more harshly simply because absolute errors there are bigger.
What is the difference between subitizing and estimation?
Subitizing is the instant, virtually effortless recognition of exact quantities up to about four items β you see three dots and simply know it is three, with no mental arithmetic required. Estimation, by contrast, is what the ANS does for larger arrays: it produces a fast but approximate magnitude impression whose precision degrades as the ratio between compared quantities shrinks. Count spans both regimes. For the smallest arrays you may subitize correctly; for larger arrays you are squarely in ANS territory. Noticing where your accuracy drops off is itself informative about your subitizing range.
Does a high score on Count mean I am good at maths?
Not directly. ANS acuity and formal mathematical ability are correlated β studies find that people with sharper approximate number sense tend to perform better on symbolic arithmetic and algebra β but they are distinct capacities. Count measures your perceptual precision for nonsymbolic quantities. Strong performance reflects a well-calibrated intuitive number sense, which researchers believe provides a foundational substrate for learning symbolic mathematics, rather than being the same skill.