Three materials, one operation

Pattern recognition test across numbers, shapes and positions

Twenty-four sequences — eight numeric, eight figural, eight tracking a cell around a four-by-four grid — shuffled together, free, untimed, with six choices on every single one so the three columns can be compared without the guessing rate getting in the way. The whole report is that comparison: your accuracy and median time in each material, each with a 95% interval, and a straight statement of whether the gap between your best and worst is bigger than the sampling noise.

100% free
No signup
24 items, 8 per material
6 choices everywhere
Interval on every column

24 items, 8 in each of three materials, shuffled so no carrier gets all the fresh attention or all the tired attention. The rule to find is the same kind of thing every time; what changes is what it is written in.

Numbers: Five numbers, state the sixth. The rule is arithmetic and the search is arithmetic.
Shapes: Five figures, state the sixth. The rule is a property changing and the search is visual.
Positions: A cell moving on a four-by-four grid, state where it lands next. The rule is a vector and the search is spatial.

All three offer six choices, which is the part that makes the comparison mean anything: if the numbers carried six options and the grids carried four, a gap between them would be partly a gap in what a guess is worth. Six everywhere, so a difference in your accuracy is a difference in you.

No clock on any item, no feedback until the end, and the number keys 1 to 6 work everywhere the mouse does.

How to compare yourself across three materials

The columns are the result; the total is almost meaningless here.

Work all three without deciding in advance which is your weak one
The materials are shuffled with no more than two of the same kind in a row, which is there so that no material gets all your fresh attention or all your tired attention. Expecting to be bad at the numbers is a good way to be bad at the numbers, and the run cannot tell that apart from actually being bad at them. Answer each item on its own terms and read the columns afterwards.
Read the three intervals before you read the three counts
Eight items give a 95% interval on accuracy about forty percentage points wide, which is wider than most gaps you will see. The results panel says outright whether your best and worst columns are separated by more than that. If they are not, the ordering is what this run happened to produce, and repeating it is the cheapest way to find out whether the gap is yours.
Use the medians to tell a weak material from a slow one
A column can be low for two different reasons and the median solve time separates them. High accuracy with a long median means the material is workable but expensive — you get there, and it costs. Low accuracy with a short median means you were not finding the rule and answered anyway. The second is the one worth acting on, and it is invisible in a single overall score.

Technical specifications

Items and materials	24 scored items: 8 numeric sequences, 8 figural series, 8 position sequences on a 4x4 grid. Interleaved with at most two consecutive items in the same material
Choice count	6 on every item in every material, which is the design decision the comparison rests on. Unequal choice counts would make part of any gap between the columns a gap in what a guess is worth
Numeric rules used	A constant step; a constant multiplier; a step that itself grows by a constant; two interleaved sequences on alternate positions; and each term the sum of the two before it. The arithmetic stays inside mental range on purpose — the work is identifying the operation, not performing it
Position rules used	A fixed row-and-column step that wraps at the edge; the same step bouncing off the edge; and a fixed number of cells along in reading order. The distractors include the previous cell and the cell one step beyond the answer, so overshooting and undershooting are both catchable
Figural rules used	A constant turn each step, a constant growth in element count, a three-step cycle on one property, and two properties on periods of two and three at once
Statistics reported	Accuracy per material with a 95% Wilson interval, median solve time over the correct items, and an efficiency figure only where a material cleared 90% accuracy — below that the divisor measures guessing rather than ability
Timing	No limit on any item. Durations are measured from the frame that painted the item to the timestamp on your answer and reported as medians over correct items only, because the time to give a wrong answer is not a measure of how fast you do the task
What leaves the page	Nothing. All twenty-four items are generated in this tab from a seed, and the copy button writes three counts, three intervals and three medians to your clipboard

Frequently asked questions

Why split it into three materials instead of giving one score?

Because one score would average away the only interesting thing in the data. Sequence rules can be carried by anything, and people are not equally fluent in every carrier — a run where you take fourteen of twenty-four could be a flat six, five, three across the three materials or a lopsided eight, four, two, and those are different facts about the same person. The average is identical and the second one tells you where to spend your practice.

Is being better at numbers than at shapes actually meaningful?

It is meaningful about you and unremarkable as a fact — most people have a clear best material, and a preference of this kind is one of the oldest observations in the study of reasoning. What it is not is a category you belong to. The results panel states plainly whether your best and worst columns are separated by more than the intervals allow, and on eight items per material the honest answer is often no. A gap that repeats across two separate runs is worth taking seriously; a gap on one run is a first look.

Are number sequences pattern recognition or just arithmetic?

Both are involved and the items here are built to keep the arithmetic subordinate. The rules are constant steps, constant multipliers, growing steps, interleaved sequences and sums of the previous two — all recognizable without paper — and where a rule multiplies, the starting values are small enough that the sixth term stays inside mental range. The point is which operation is running, not whether you can execute it. If the execution is what is hard for you, the numeric column here is measuring that instead, and the comparison against the other two columns will say so by being lopsided for the wrong reason.

What if reading digits or seeing fine detail is difficult for me?

Then the numeric column is measuring reading and the comparison across the three is measuring your reading against your reasoning, which is not what this page claims to do. The same is true in the other direction for anyone whose difficulty is with visual detail: the figural items depend on telling shading and orientation apart at a small size. The two other columns still stand on their own, and reading the run as two materials rather than three is the honest way to use it. Nothing on this page attempts to detect any of that, and no result here identifies a condition.

Do the grid items measure spatial ability?

They measure serial rule detection with a spatial carrier, which is a narrower thing than spatial ability as it is usually meant. Rotating an object in your head, judging what a folded sheet looks like or reading a map are different tasks with their own histories, and none of them is here. What the grid items do share with those tasks is that the rule is a vector rather than a value — the answer is where something goes rather than what it becomes — and that alone is enough to separate people who found the numeric rules effortless.

Why is there no combined speed-and-accuracy score?

There is, and it is withheld unless the column earns it. Dividing mean time by proportion correct folds both into one figure and is standard practice, but it breaks down below roughly ninety percent accuracy, where the divisor stops describing ability and starts describing how much of the score was guessing — at which point a fast guesser scores best. The panel shows the figure for any material where you cleared that bar and explains its absence for the ones where you did not, rather than printing a number that flatters the wrong behavior.

Can I take just one of the three materials?

Not on this page, because a column on its own is exactly the thing the page is designed not to produce: eight items with no comparison is a number with a forty-point interval and nothing to hold it against. If you want figural sequences alone with the rule-discovery cost timed, that is a different measurement on a different page in this set. What you can do here is run the whole twenty-four again — the items are regenerated from a fresh seed each time.

Why the same rule is easier in one material than another

“Pattern recognition” is a folk term covering at least four different operations — spotting a repeat, matching a template, classifying a new instance, and recovering the rule behind a sequence. This page does only the last one, and does it three times in three materials, because that is where the interesting variance is. When factor-analytic work on human abilities settled into its modern shape, the organizing axis it kept finding underneath reasoning was content: figural, numeric and verbal material behave as partly separate abilities, not as one ability wearing three coats. Carroll’s 1993 survey of the whole factor-analytic literature is where that is laid out most fully, and it is the reason a single pattern-recognition score is a weaker statement than three.

The design decision that makes a three-way comparison honest is boring and almost always skipped: the same number of answer choices in every material. If the numeric items offered six options and the grid items four, then a visitor guessing on both would score higher on the grids, and part of every gap between the columns would be the guessing rate rather than the person. Six everywhere costs the figural items a distractor they could have supported and costs the grid items nothing, and it buys the only thing that makes the columns comparable. The second decision is interleaving: eight items of one material in a block would put one column at the start of the run and one at the end, and whichever material came last would be carrying however tired you were by then. The third is the interval on each column, which on eight items is wide enough to be embarrassing and is printed anyway, because a page that shows three counts without them is inviting a reading the data cannot support.

There is no population figure to sit beside any of this, and the reason is the same one that applies to every generic reasoning format: an average would describe the difficulty this page chose for its own items rather than anything about people, and changing one rule in the numeric set would move it. The comparison that survives is you against yourself across three materials on the same day. For the figural column broken down rule by rule there is the matrix reasoning diagnostic, and for the cost of finding a figural rule the first time, the inductive reasoning test. Cross-material effects of a much stranger kind — a digit that reliably has a color, a letter that has a position in space — are what the synesthesia test measures by consistency rather than by accuracy, and if you are here because a recruiter mentioned pattern recognition, the questionnaire arriving alongside the reasoning section is usually a personality assessment, which is scored on nothing like these principles.

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.

Because the three medians are compared against each other rather than against an outside figure, that floor sits under all three equally and does not move the comparison. It would matter if this page claimed a material was faster by twenty milliseconds; it claims differences in seconds, and says so.

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 dyscalculia, a visual processing difficulty or a spatial learning disability. Only a qualified professional, working with more than a browser, can make that judgment.

Where the twenty-four sequences are generated

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 numeric sequences, the figures and the grid paths are all computed in this tab from a single random seed, so the page ships no item bank and downloads nothing when you press start. The three columns are worked out from the same tab’s memory and disappear with it.