Boxes that do not say what they do
Diagrammatic reasoning test where the operator is the unknown
Fifteen flows, free, no clock and no signup. In each one a symbol travels into a lettered box and something comes out the other side, and the box carries no description of itself — what it does has to be recovered from the examples printed above the flow. Five choices an item throughout, six boxes in the whole vocabulary, and a report that names the box you kept misreading rather than handing you a bare count.
- 100% free
- No signup
- 15 flows, 3 question forms
- 6 unlabeled operators
- Error kinds named
15 flows. A symbol goes into a lettered box and comes out changed; the box never says what it does, and working that out from examples is the whole task. Five choices an item, no clock, and three different questions asked about the same six boxes.
One worked all the way through
Box A does one thing: one more element, back to one after four. The two examples show it on two different symbols so the change is separable from the symbol, which is the only reason two examples are given rather than one.
The three questions
- Work out the box, then use it. Two examples of one box, then a new symbol through it.
- Two boxes in a row. One example each, then a symbol through both. Two of the wrong answers are each box applied on its own, so leaving a step out is caught rather than half-credited.
- Name the box. A symbol and what came out, with five candidate boxes demonstrated on other symbols. Nothing can be matched by appearance here, because no candidate is shown acting on the symbol in the question.
The worked example above uses a box you will meet again, which is deliberate: the six boxes are a fixed vocabulary and knowing it is not the test. Which box is in this particular flow is.
How to work a flow you have never seen
One habit does most of the work here, and one habit ruins the naming items.
Compare the two examples with each other before you look at the question
A single before-and-after pair is consistent with more hypotheses than it looks: a triangle that came out pointing sideways might be a quarter turn, or it might be a rule about triangles specifically. The second example is on a different symbol, and everything that survives both is the box. Reading the examples one at a time and then guessing from the second one is the most common way to lose a flow you understood.
On a two-box flow, draw the middle symbol before you look at the choices
The wrong answers are built from exactly the mistakes that skipping the middle produces — each box applied on its own, and the last box applied once too often. If you know what came out of the first box, none of those is tempting. If you go straight from the input to the answer strip, all three are, because each of them is a real symbol that a real reading of the flow would produce.
On a naming item, describe the change in words rather than looking for the picture
The five candidate boxes are demonstrated on symbols that are not the one in the question, deliberately, so there is nothing on screen to match against. What transfers is a sentence: it grew, it turned, it gained a mark, it became the next shape. Once the change is in words the candidate that produced the same sentence is quick to find, and the ones that changed some other property drop out at a glance.
Technical specifications
| Flows and question forms | 15 scored flows in three forms of five: two examples of one box then a new symbol through it; one example of each of two boxes then a symbol through both; and an input with its output, to be matched against five candidate boxes. Shuffled with at most two of a form consecutively |
|---|---|
| The operator vocabulary | Six boxes, each acting on a different property of the symbol: its orientation, its shading, its size, how many elements it has, the mark sitting above it, and which shape it is. Every one is cyclic, so it always produces a visible change and never runs out |
| Why two examples and not one | One example leaves the rule under-determined, because a change seen on a single symbol cannot be separated from that symbol. Two examples on different symbols rule out every hypothesis tied to the particular figure, which is the smallest number that makes the item well posed |
| Order in the two-box flows | The two boxes in a chain always act on different properties, so they commute and the answer does not depend on which runs first. The load is holding the intermediate symbol, not sequencing, and the wrong answers are built as each box applied alone rather than as a reversal |
| Answer choices | 5 on every item in every form. A blind guess is worth 0.2 of a flow, so guessing all fifteen averages 3.0 correct, which is the figure your total is tested against with an exact binomial probability |
| Wrong answers are categorized | Every distractor on a flow item is one named error: the input untouched, the box applied one time too many, only the first box, only the second box, or a different box altogether. The results panel counts how many of each you picked |
| Timing | Nothing is timed out and nothing is cut short. Per-flow durations are recorded from the frame that painted the flow and reported only as a median for each of the three forms, since the interesting comparison is between forms rather than against a clock |
| What leaves the page | Nothing. The flows, the symbols and the distractors are generated in this tab from one seed; the copy button writes the form table, the box table and the error counts to your clipboard |
Frequently asked questions
Why does each box come with worked examples instead of a legend?
Because a legend would turn the item into a lookup. If the page printed 'box A rotates the symbol', the only thing left to do is rotate a symbol, which is a drawing exercise. Recovering what the box does from cases of it working is the whole measurement, and it is the reason this format is used at all: an unlabeled operator is a function you have to infer, and the inference is the part that varies between people.
The naming items feel much harder than the others. Is that intended?
Yes, and the reason is structural rather than a matter of difficulty tuning. On the other two forms you build a result and look for it among the choices, so a correct answer can be recognized. On a naming item there is nothing to recognize — the candidates are demonstrated on symbols unrelated to the one in the question, so the comparison has to be made between two descriptions of a change rather than between two pictures. That step is exactly the abstraction the format is named after, and it is where the run separates people who worked the earlier flows by appearance.
Does knowing all six boxes in advance defeat the test?
It changes what is being measured and it does not remove it. The vocabulary is deliberately small and the worked example on the start screen uses a box you will meet again, because remembering six operators is a memory task and not an interesting one. What stays hard once you know the menu is deciding which box is in front of you from two examples, holding an intermediate result through a chain, and matching a change against candidates you cannot see acting on your symbol. On a second run the operators are the same and every symbol, pairing and distractor set is new.
Is this the test recruiters send out under a different name?
Probably the same family, under one of several names. Graduate assessment publishers use this format for engineering, software and analyst roles and label it diagrammatic, logical, symbolic or inductive depending on the publisher, usually with a strict clock and a fixed item bank. This page is not any of those instruments, is not affiliated with any of them, and is not scored against their norms. What it shares with them is the item structure, which is what makes it useful as practice.
Should I write down what each box does as I go?
It will not help as much as it feels like it should, because the letters are local to a flow. Box A in one flow and box A in the next are unrelated draws from the same six, so a written key from three flows ago is worse than useless — it is misleading. Notes made inside a single flow are a different matter, particularly the middle symbol on a chain, and the run is untimed precisely so that working on paper is not punished.
Why does the box table add up to more than fifteen?
Because a two-box flow is counted under both of its boxes. Five chains contribute two rows each, so the six operator rows sum to twenty across a full run rather than fifteen, and a single flow can put a mark against two boxes at once. That is the price of getting a per-operator breakdown from only fifteen items, and it means a low row is a hint about which change you find hard to see rather than a measurement of it.
What an unlabeled operator asks for that a sequence does not
A series item asks what comes next. A matrix asks what fills the hole. A flow asks something structurally different: the unknown is not a term but a function, and the only evidence for it is a handful of cases of that function doing its job. That framing is older than the assessment industry that uses it — a transformation, in the sense Ashby set out in his 1956 Introduction to Cybernetics, is defined by what it does to every operand it can accept, and not by any name attached to it. A lettered box on a graduate assessment is that definition drawn as a picture, and reading one means reconstructing the whole mapping from two instances of it.
That is also why the items here give two examples rather than one, which is the detail most practice material gets wrong. Show a triangle going in and a rotated triangle coming out, and the honest set of hypotheses includes the quarter turn, a rule that flips triangles, and a rule that points every symbol in one particular direction — all consistent with the evidence on offer. A second example on a different symbol kills every hypothesis that was really about the first symbol. Two is the smallest number that makes the item well posed, and any distractor set built on top of a badly posed item punishes people for hypotheses the item never excluded. The three question forms then load different things on purpose: working out a box and using it is inference followed by application, a chain adds an intermediate result that has to be held while the second box is applied, and naming a box removes appearance matching entirely by demonstrating every candidate on symbols other than the one in front of you.
No average exists to put beside fifteen flows, and the reason is worth stating plainly rather than hiding behind a missing table. This page chose its own operators, its own symbol properties and its own distractors, and every one of those choices moves the score; an average produced from them would describe the choices rather than the people who took the run. What the report does instead is internal: which of the six boxes you misread, whether chains cost you more than single boxes, and which of the five named errors your wrong answers were. The timed, mixed-format version of this material sits on the abstract reasoning test, the search that happens before any operator can be applied is isolated on the inductive reasoning test, and the same inference over physical rather than symbolic systems — levers, gears, pulleys — is the mechanical reasoning test. Two neighbours are worth knowing about for the opposite reason: an answer that arrives before any box has been worked out is what the cognitive reflection test is built to catch, and the same wariness about a conclusion that merely sounds like it follows, applied to arguments written in English rather than drawn in boxes, is the critical reasoning page in this set.
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.
Only the three medians on this page carry it at all, and they are reported to a tenth of a second because a flow takes tens of seconds to work. Nothing in the scoring depends on a duration: the counts, the intervals and the box table are all built from which option you picked.
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 an aptitude for technical work, a fit for a particular role, or a difficulty with abstract thought. Only a qualified professional, working with more than a browser, can make that judgment.
Where the fifteen flows are built
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 symbols, the operators behind each box and every wrong answer are computed in this tab from a single random seed at the moment you press start, so there is no item bank to download and nothing to fetch between flows. Your choices exist in this tab’s memory and are gone when it closes.