What this test measures
A number appears, stays on screen for 0.9 seconds plus 0.2 seconds per digit, then vanishes. You type back what you saw. The first number is three digits long, and every correct answer makes the next one a digit longer. The first wrong answer ends the run. The screen then shows what you typed next to the real number, so you can see exactly where it fell apart. Your score is the number of digits in the last one you got right. An optional hard mode shows each number for only 60 percent of that time, and keeps a separate record.
The ability being exercised is short-term memory for a verbal sequence, what psychologists call digit span. Most people hold digits by silently repeating them. It is the same trick you use to keep a phone number alive between reading it and dialling it. Span is a capacity measure: how much ordered material you can hold for a few seconds before it decays. It is not intelligence. A strong run does not mean you are a good thinker. A weak one says nothing about how you learn, or how well you remember your own life.
Why there is no percentile here
Finish a run on this site and you get a digit count — no percentile. That is deliberate. We only attach a percentile when a published study truly matches the task. For this test, we could not find one that does.
Digit span norms exist; the problem is fit. The well-matched ones we could actually verify come from specific populations. Iñesta and colleagues (2021), for example, report a mean forward span of 5.36 digits (SD 1.15) in a Spanish sample aged 55 to 87. Grading a twenty-five-year-old against a table built from retirees would produce flattering percentiles that mean nothing. The classic clinical version adds a second mismatch: an examiner reads the digits aloud at one digit per second and you answer out loud. Here the whole number sits on a screen and you type it back. Reading is not listening, and typing is not speaking. Score a slow typist against those norms and you punish their hands, not their memory. Even the most famous benchmark in the field, the "seven items" figure from George Miller's 1956 paper, came from read-aloud experiments — never from a screen test like this one.
We could paper over all of this with a plausible-looking curve. We would rather show nothing than a number we cannot defend, so until someone publishes norms for this exact format, the percentile stays blank.
The "seven items" myth
Miller's paper, "The Magical Number Seven, Plus or Minus Two" (Psychological Review, 1956, 63(2), 81-97), gave the world its favorite memory number, and the number escaped the lab almost at once. Seven digits, seven list items, seven of anything became the assumed ceiling of human short-term memory. Miller himself was partly joking. The paper opens with a joke: the number seven keeps following him through his data. And most of it is really about something else — how repackaging information changes how much a person can hold.
Later research treats seven as an overestimate for pure short-term storage. When experiments strip away rehearsal and grouping, the capacity that remains looks closer to three to five items. People reach seven and beyond because they quietly repeat the material and pack it into larger units, not because the raw store is that big. Your score here reflects raw storage plus whatever strategy you brought to it, which is exactly why the next section matters.
Chunking: how people beat their own limit
Try to hold 1 7 7 6 2 0 0 1 as eight separate digits and you will probably drop one. Read them as 1776 and 2001, two familiar years, and the load falls to two items. The digits did not change; the packaging did. That is chunking, and it is the main reason practiced people post scores far above the textbook range.
Phone numbers are the everyday example. Nobody stores 8005551234 as ten separate digits. The formatting splits it into three short groups, and each group rides along as a single unit. Dates do the same job. So do ages: 42 is easier to keep as "my mother's age" than as a four and a two. The richer your stock of hooks, the more digits each chunk can absorb.
The best-known demonstration of this involved a college student, a distance runner, who practiced digit recall over months of laboratory sessions. He started turning digit groups into running times he knew. Then he built bigger groupings on top. In the end his span grew to many times the normal range. Two details from that study matter here. His improvement came from encoding, layer upon layer, not from his memory physically expanding. And when the experimenters switched from digits to letters, where his running-time trick was useless, he fell straight back to an ordinary span.
What changes your score
Span is not a fixed constant you carry around. On this test, several things move it, and some of them have nothing to do with memory.
Rehearsal sets the first limit. Most people keep the digits alive with inner speech, so how quickly you can cycle through them silently caps what you can hold, and digits that sound alike blur together. Distraction eats into the same resource. Background speech is especially damaging because it competes with the inner voice doing the rehearsing, and a quiet room helps more than people expect.
Typing is the second limit, and the test cannot tell a memory failure from a finger failure. Hit an adjacent key on a twelve-digit answer and the run ends the same way, with the score under-reporting what you actually held.
Display time is fixed. The number shows for 0.9 seconds plus 0.2 seconds per digit, and hard mode cuts that to 60 percent. Longer numbers get more time, but you cannot pause or stretch it, so whatever reading and grouping you do has to fit inside that window. On top of all this, sleep, fatigue, stress and time of day shift performance from one session to the next, which is one reason repeat runs disagree.