Attention

Can you hold a rule while something fights it?

An arrow points one way. Written across it is a word that often says something else. Answer the arrow. Ignore the word. Forty-eight times, with a deadline on every one.

The word is the point. Reading is so automatic that you cannot switch it off, so on most trials part of you has already answered before you have looked at the arrow — and your job is to not let that part win.

Before you start

It takes about two minutes, including eight practice trials that are not scored.
Each trial has a deadline of 0.65 seconds. If you have not answered by then the trial closes and counts as wrong. That is deliberate and the reason is published below.
Use the arrow keys if you have them. There are four on-screen buttons as well, but a keyboard is easier at this pace.
You will get some wrong. Everybody does — the task is built so that you do. A run with no errors usually means the deadline was not reached, not that you are perfect.
This task cannot be done without seeing it, and that is a real exclusion rather than an oversight. Each trial is announced in a live region, but a spoken announcement of the arrow and the word takes longer than the 0.65-second deadline it is announcing — so the deadline would expire during the reading of it. Every other instrument on this site states whether it works without sight; this is the first one where the honest answer is no.
Nothing is stored. No account, no cookie, no record. Close the tab and the result is gone.

This test is one of the seven arms of the Cognitive Fingerprint. It was not, until v2.0 — the Fingerprint shipped with four arms and arms were added in one deliberate version change once the domain set was complete, rather than one at a time, because the scores in a shared link travel in the link and adding an arm changes the answer a sent link resolves to. A link made when there were four arms is still read on the four-arm scale and keeps its number. How the number is built, and what the version marker does.

Your result

—

What that says about the task

Where the errors fell

What this task tested

Holding one rule in mind — answer the arrow — while a second, faster, unasked-for process supplies a different answer. That is the thing the score is about, and the deadline is what makes it visible.

Reading is automatic in a way that looking at an arrow is not. When the word and the arrow agree, the automatic answer is the right one and you get it for free. When they disagree, you have to override an answer you did not choose to produce. The score is how often you managed that at a pace fast enough for it to be hard.

What this did not measure

How fast you are. Reaction time is not in the score anywhere. This is the only cognitive measure on this site where that is a refusal rather than an oversight — see the deadline section.
Attention in the everyday sense. Whether you can sit with a long document, resist your phone, or stay with a boring meeting. Those are largely about motivation, environment and habit. This is a two-minute laboratory-shaped task and it does not reach them.
Anything clinical. This is not a screen for ADHD or for any other condition, it cannot be used as one, and a low score here is not evidence of one.
Your attention at any other moment. One run, one state, one device.

Things that moved this number that are not your attention

Your device. The deadline is 0.65 seconds from the moment the trial is drawn, and the drawing itself takes time we have not measured on your hardware. A slower screen makes this task harder. That is a real limitation and it is not correctable from here.
Whether you used a keyboard. Arrow keys are faster than aiming at a button. Reaching for a button with a thumb can cost more than the whole margin.
Practice. The eight practice trials help, and familiarity with the task keeps helping. Retest gains on cognitive tests average around a fifth of a standard deviation from a first sitting to a second and plateau after the third — that is an average across studies, not a prediction about you, and it is a reason to read a second run cautiously rather than a reason to expect a particular one.
Whether you were interrupted. At this pace a two-second distraction costs three trials.

Should you believe this number?

As a description of how this run went: yes. Forty-eight trials, a fixed rule, a fixed deadline, and a count of the ones you got right. There is no modelling between what you did and what is printed.

As a statement about you: much less. It is one run on one device in one state, and we have no reference sample — Rung 0, reference sample zero. Nothing on this page compares you to anybody, because we are not entitled to.

The weakest thing about it, stated plainly: the 0.65-second deadline is a number we chose. It is reasoned from a published finding and the reasoning is on this page, but it was not calibrated against measurements of our own, and a different deadline would produce a different score for the same person. Why two attention tests disagree goes into this properly.

An online test cannot tell you whether something is medically wrong. This one cannot, and neither can we. That is not caution — it is the limit of the task. If your attention or your concentration is worrying you, a doctor is the right person to ask. And you do not need a low score here to justify the appointment.

Two questions, if you have twenty seconds

Nobody is asked to sign anything and no text box is offered. Two taps.

Did the page make it clear what this score is and is not?

What next

Why can't you just ignore the word? — what Stroop found in 1935, and why eight days of practice shrank the effect without removing it.
Why two attention tests give you different answers — including ours against the one this task is usually compared with.
Every instrument on this site, with what each one measures and what each refuses to claim.
Why there is one number and what it is made of.

How this is scored

Everything below is the whole rule. There is nothing behind it.

The task

Forty-eight scored trials, preceded by eight practice trials that are discarded. On each trial an arrow is drawn pointing up, down, left or right, and a direction word is written across it. The answer is always the arrow.

24 trials are congruent — the word matches the arrow.
24 trials are incongruent — the word is one of the other three directions, chosen at random from those three.
Each of the four directions is the answer exactly 12 times, and each of the four words appears exactly 12 times. The order is shuffled.
Every trial has the same deadline: 650 milliseconds from the moment it is drawn. A trial with no answer by then closes and is scored wrong.

The deadline, and why the score is not a speed

An untimed version of this task measures nothing. That is not a guess. In a published study of exactly this family of tasks, congruency effects on accuracy "are found only in trials with fast responses (< 600 ms)", with accuracy "attaining near-perfect levels at about 600 ms" (Gratton and colleagues' conditional accuracy analysis of the Eriksen flanker and spatial conflict tasks — source below). Give people as long as they like and almost everyone scores almost everything correctly, and the test has told you nothing.

So the deadline is not decoration. It is the thing that turns accuracy into a measurement at all, and 650 milliseconds is set just above the band where that source finds accuracy still discriminates, with a small margin for the browser's own delay. [S] The margin is an inference, not a measurement — we have not measured display and input latency on your device, and we say so rather than implying a precision we do not have.

The score is a count of correct answers and nothing else. The obvious alternative — the standard one — is to subtract your incongruent reaction time from your congruent reaction time and report the difference. We refuse that for two reasons, and the first is the serious one:

A difference score is not permitted as a domain output here (D-062). Two noisy measurements subtracted from each other produce a third that is noisier than both, and it is the kind of number that looks precise and is not.
And the effect is tens of milliseconds. We have published our own measurement of a touchscreen adding four times more latency than the response it was supposed to be measuring. In a browser, that difference is mostly hardware.

The scoring rule

Your score is the number of the 48 scored trials on which you pressed the direction the arrow pointed, before the deadline. There is no weighting, no bonus for the hard trials, and no penalty for errors beyond not getting the point.

Ceiling: 48. Unlike the ninety-second arithmetic ceiling, this one is not a choice — it is how many trials there are.
Pressing at random scores about 12. Four options, so one guess in four is right. Verified by simulation rather than asserted: over 200,000 generated runs answered at random, the mean was 11.99 of 48.
Not answering at all scores 0, which is below chance, because a missed deadline counts as wrong.

Exclusion rules — when an attempt does not count

More than 8 of the 48 trials closing with no key pressed at all. That is an interruption or a stalled device, not a measurement, and no band is given.
A score of 12 or below. At or under what random pressing gets, the run says nothing about you and is not banded.
The tab being hidden at any point during the run — switching tabs, switching apps, minimising the window. The run is voided, with no exceptions and no partial credit. This is stricter than it sounds and it is not tidiness: when this page is not the visible tab, the browser stops its timers almost completely. Measured on this page on 7 September 2026: zero trials advanced in twelve and a half seconds, against a design of about one per second. The deadline quietly becomes no deadline, and since the deadline is the entire measurement here, the run would produce a high number that means nothing at all.
Reloading or leaving the page mid-test. The run is discarded, not resumed.
The practice trials are never scored, in any circumstance.

The bands

The bands describe the task, not you, and not other people. No band on this page is a percentile or a comparison, because at Rung 0 with a reference sample of zero we are not entitled to make one.

44–48 — you held the rule almost every time at a pace built to break it.
36–43 — the rule held most of the time, and the word won occasionally.
25–35 — the rule held more often than not, and the deadline was doing real work.
13–24 — above what guessing gets, but the pace was faster than the rule could survive.
0–12 — at or below random pressing. No band. Something went wrong with the run rather than with you.

What we do with your answers

Nothing leaves your browser except, if you press one of the two feedback buttons, a single count on our side. No score, no trial data, no identifier, no account, no cookie. See Privacy.

Sources

Gratton, G. and colleagues, Conditional accuracy in response interference tasks: evidence from the Eriksen flanker task and the spatial conflict task — the finding that congruency affects accuracy only in fast responses, which is what forces the deadline. Read at source, 7 September 2026.
Fan, J., McCandliss, B. D., Sommer, T., Raz, A. and Posner, M. I. (2002), Testing the efficiency and independence of attentional networks — the canonical attention paradigm, which reports its results as reaction-time difference scores and is the thing this instrument deliberately does not do. Abstract read at source; the full text was not read, and that is recorded in our evidence ledger as a partial verification.
Our usual anchor does not cover this domain. The ETS Kit of Factor-Referenced Cognitive Tests — the source behind four of our other instruments — has no attention or concentration factor among its 23 cognitive factors. We say so rather than implying a lineage this test does not have.