Rare targets, held rule

Continuous performance test, block by block

Letters arrive one every 1,750 ms for ten and a half minutes and you press the spacebar only for an X that an A came immediately before — 13.3% of them qualify, so most of this is waiting, which is the design. The run reports the targets you missed, the three different wrong presses counted against three different denominators, how far your hit times wandered, and what changed between the first block and the last. It is free, the numbers appear the moment the run ends, and nothing is uploaded. It is not a clinical continuous performance test, it produces no cutoff, and it cannot contribute to an assessment of anything.

100% free
No signup
360 letters
10 m 30 s
Four scored blocks

Letters arrive one at a time, each on screen for 250 ms with 1,500 ms of blank after it. Press Space for an X — but only when the letter immediately before it was an A. Every other letter, including an X that arrived on its own, wants nothing from you.

13.3% of the letters qualify, so most of this is waiting. That is the design and not a flaw in it: what a rare target measures is whether the rule is still being applied after ten minutes of it not mattering.

The stream, spelled out

Rule: press only for X when the letter immediately before it was A
Stream: 4 blocks of 90 letters — 360 in all, 10 m 30 s
Targets: 12 per block, 13.3% of the letters
Decoys: 12 Xs per block with no A in front, and 12 As followed by something else
Timing: 250 ms on screen, 1,500 ms of blank, one letter every 1,750 ms
Window: 1500 ms from the letter appearing; the blank is still inside it
Interruption: this window losing the front ends the run rather than pausing it, and what was collected first is still scored

Every A in the stream is built as the first half of a two-letter unit, so no A can land next to an X by accident where the shuffle happened to join two units. Without that rule the sequence would contain targets the design never counted, and a correct press on one would be scored as an error.

One block gives you 12 targets and no comparison between blocks at all, which is to say every figure except the one this page is for. Take it to find out whether you can sit through the other three.

How to sit the continuous performance test

One rule that needs two letters, and ten minutes of it mostly not mattering.

Hold the rule: X counts only when A came just before it
An X on its own wants nothing from you. An A on its own wants nothing from you. Only the pair, in that order, with nothing in between, is a target — which means the letter on screen is never enough on its own and you are always carrying what came before it. Rest one finger on the spacebar and leave it there. The wide bar under the panel takes a tap instead if you have no keyboard, and the result counts how many of your presses came that way, because a thumb that has to travel is a thumb that was not already poised on a rare target.
Take the thirty practice letters, because the errors are the interesting part
Practice runs on a wider window and names each trial back to you afterwards, which is where you find out that an X arriving without its A is genuinely hard to leave alone. None of it reaches any figure. It is worth taking even if the rule seems obvious: the decoys are half the design, and meeting them for the first time inside the scored run costs you a block.
Sit the four blocks with the tab in front, and read the last against the first
A break comes after each block of 90 letters. Leaving the tab does not pause this run, it ends it — a stretch of attention with a gap in it is not what was being measured, and padding the missing letters back on would produce a complete-looking block whose decrement figure is fiction. What was collected before the break is still scored. At the end you get a table with one row per block: targets found, wrong presses, hit time and its spread, plus the difference between the first block and the last with a 95% interval on it.

Technical specifications

Length	4 blocks of 90 letters, 360 scored trials, 10 minutes 30 seconds of stream plus three breaks you control. A single-block option runs 2 minutes 38 seconds and gives every figure except the one across blocks
Target	X immediately preceded by A. 12 per block, 48 in the full run — 13.3% of the letters, so the rule matters on roughly one trial in eight
Decoys	12 X's per block with no A in front of them, and 12 A's per block followed by something other than X. Both are counted separately from each other and from the 18 letters per block in neither role
Timing	Each letter is on screen 250 ms, followed by 1,500 ms of blank, one letter starting every 1,750 ms. The interval is held by setting each wait from how long the previous trial actually consumed, so a press does not pull the next letter forward
Response window	1,500 ms from the letter painting, which continues through the blank after the letter has gone. Anything later belongs to no trial and is counted as a stray press rather than charged to the letter that follows
Alphabet	A, X and ten fillers — B, C, D, E, F, G, H, J, L, T — in a monospaced face on a dark panel. Every A in the stream is built as the first half of a two-letter unit, so no A can end up beside an X where the shuffle joined two units
Interruption	This window losing focus ends the run instead of pausing it, unlike every other timed page on this site. Whatever was collected before that point is still scored and the results panel says the run was cut short
Reference figures	No norm and no cutoff. One figure appears as context only: watch-keeping performance takes roughly 30 minutes to fall measurably (Mackworth, 1948), which is three times longer than this run

Frequently asked questions

Is this the same test a clinician would use to assess ADHD?

No. The instruments used that way are commercial products sold with software that fixes their stimulus timing to the millisecond and ships norm tables calibrated to that exact timing, and even there the result is one input among several in an assessment carried out by a person. This page runs the same paradigm with its own letters, its own target share and its own interval, and it therefore has nothing to compare your numbers against and offers no threshold to fall below. What it can do is show you the shape of the task and give you your own figures with the design parameters attached.

Why does an X on its own not count? I kept pressing on those.

That is the point of the design, and the presses you are describing are counted separately for exactly that reason. If X alone were the target, the task would only ask whether you can see a letter — the rule would live entirely in what is currently on screen. Requiring an A first means you have to carry something across 1,750 ms of other letters, and the errors then tell you which part failed: pressing on a bare X is responding to the target's appearance instead of to the sequence, while pressing on the A itself is holding the rule and firing a letter early. Those are different failures, and a page reporting one combined error rate cannot tell them apart.

What does my hit-time variability mean, and why is it printed as prominently as my speed?

It is the more informative of the two on a task this slow. A vigil almost never fails by everything getting uniformly slower; it fails by producing the occasional very long response while the rest stay ordinary, and a mean absorbs that while a standard deviation shows it. The page also prints the spread relative to the mean, so a run whose average time is high for hardware reasons is not automatically read as a wandering one. What none of this supports is a comparison against anybody else, because the level depends on your display and your keyboard and the spread depends on both plus the length you sat.

Ten minutes is a long time. What do I lose by taking the single-block option?

Everything that needs a comparison between blocks, which is the figure this page exists for. One block gives you 12 targets, so a miss rate with a very wide interval, three error counts and a hit time — all real, none of them about sustained attention, because sustained attention is by definition a claim about what changed over time. The single block is there so you can find out whether the task is legible on your screen and whether you can sit still for it, and the results panel says which figures it could not form.

The tab went behind something and the whole run stopped. Why not just resume?

Because the run would then be a lie about its own length. Every other timed page here voids the interrupted trial and puts it back on the end, which is right when the measurement is a per-trial reaction time — the trials are interchangeable. Here the measurement is what happened across ten unbroken minutes, and forty seconds spent in another window is not a gap in the data, it is a rest in the middle of a vigil. A block reconstructed around that rest would produce a decrement figure that describes the interruption. So the run ends, says so, and scores what it actually saw.

My last block was faster than my first. Should it not get worse?

Not necessarily over ten minutes, and this is the most common misreading of a vigilance run. Two things move in opposite directions across a session: practice, which speeds you up over the first few minutes, and time on task, which eventually slows you down. In a ten-minute run they are the same order of size, so the difference between the first and last block routinely comes back spanning zero or slightly negative. The classic decline was measured over a two-hour watch and had not clearly appeared in the first half hour, which is why the reference figure on this page is a caution against reading too much into a short run rather than a benchmark to beat.

Why does a press during the blank sometimes void the next letter?

Because the engine has to attribute every press to some trial, and the trial that owns the moment a press lands is the one it is inside. A press that arrives more than 1,500 ms after its letter has missed its own window, and the next trial has already begun its 250 ms of fixation — so it is recorded as a press that arrived before a letter had painted, and that trial is voided rather than scored as an error against a letter you never saw. The count of those appears in the results panel. On a fast task the situation almost never comes up; on this one, where nine trials in ten are answered by doing nothing, a genuinely late press has somewhere to land.

What the A-X form adds, and why no rate here can be compared to a published one

A continuous performance test is a long stream of stimuli with a rare target in it, and the family covers several different rules. The version here is the A-X form, in which neither letter is a target by itself: the X qualifies only when an A came immediately before it. That structure is what makes the task about more than watching. Between the cue and the letter that might justify it there is a second and a half of blank and, often, several other letters — so the rule is held across an interval, then either used or thrown away, dozens of times a minute, for ten minutes. A simpler form where the target is a single letter measures whether you are still looking at the screen. This one measures whether you are still applying a rule to what you find there, and the errors it produces separate into kinds instead of piling into one rate.

Those kinds are the reason the results panel has three rows where most versions have one. A press on an X with no A in front of it is the target’s shape overriding the sequence. A press on the A itself is the sequence being held and acted on one letter early. A press on a filler is neither, and is usually a hand keeping time with the stream. They are counted against separate denominators — 48 lure X’s, 48 unfulfilled cues, 72 fillers in a full run — because a rate is meaningless without the number of opportunities that produced it. The same principle governs the headline miss rate: it is printed next to the target share and the interval between letters every time it appears, since changing either of those moves the same person’s rate across most of its available range.

Why no reference figure sits beside your rates

The clinical instruments in this space are commercial products whose norms are sold with the software and are specific to their exact stimulus timing. Their scores are also routinely used in diagnostic workups, which is the zone this site does not enter.

One figure does appear, and it is a caution rather than a comparison. The measured deterioration that gives this whole family of tasks its purpose takes far longer to arrive than any web page can hold somebody:

Onset of the vigilance decrement

≈ 30 minutes

Mackworth (1948), The breakdown of vigilance during prolonged visual search, Quarterly Journal of Experimental Psychology

The original watch ran two hours on a moving-pointer clock, and the decrement's timing depends on how often signals arrive and how obvious they are — a rare, faint signal degrades attention faster than a frequent, clear one. A browser task of three or five minutes ends before the phenomenon starts, so a short test cannot report sustained attention and should say so rather than implying it measured it.

So ten and a half minutes is an honest compromise and is labeled as one: long enough that four blocks can be compared and boredom is genuinely part of the experience, short enough that people finish. If you want the same restraint measured without the held rule, the go / no-go test makes the current stimulus decide everything and asks only whether the hand can be stopped. For the drift of ordinary speed across time with nothing rare in it at all, the attention span test answers every trial instead of withholding most of them, and the concentration test replaces the waiting with a clock and a dense sheet. Interference rather than vigilance is the Stroop test and the flanker task, picking a target out of clutter is the selective attention test, and if what interests you is how confidently the eye reports things that are not there, the optical illusion test is the other end of the same question.

A reaction time here is the interval between the frame that painted the stimulus and the timestamp the browser attached to your key, both read from the same monotonic clock. What neither can see is the display pipeline behind it, so on a 60 Hz screen roughly 16 ms of every figure below is the machine rather than you. That is the timing floor: two numbers closer together than that are the same number, and this page reports no precision it cannot support.

This is a measurement exercise, not a clinical assessment. It reports what you did on this page against a stated reference and nothing more — it cannot establish ADHD, an attention disorder, or any condition at all. Only a qualified professional, working with more than a browser, can make that judgment.

Where ten minutes of letters and presses end up

Every number on this page is worked out by JavaScript running in the tab you are reading it in. Your answers, your reaction times and your score are never uploaded, logged or kept — which is also why the test carries on working after you disconnect from the network, and why nothing here can be held back behind an email address.

The letter stream is generated in this tab from a seed printed with your result, so the exact 360 letters you saw are reproducible from that number and never have to be kept. Your press times live in one array until the tab closes; a reload loses the run, which is why the copy button carries every block rather than a summary.