Transparency
How our questions are made
You are preparing for an exam that decides a real job, that pays real money. You deserve to know where the questions come from, how they are built and how the system ensures quality.
Abstract reasoning
An abstract question starts as a rule, not as a picture. A rule can make only these moves:
Turn by a fixed step
Next shape in a cycle
One size up or down
One shade lighter or darker
One more or one fewer
One place around a ring
Quarter turn or mirror flip
One to three of them run at once. The rule is applied five times to draw the figures you see and a sixth time to produce the answer, so the answer is not a matter of opinion.
One rule, applied six times. You are shown the first five, and the sixth is the answer.
- The four wrong options break that rule in one named way: the series carried one step too far, the previous figure repeated, the step doubled, one rule frozen while the others move on, or one rule run backwards.
- Before a question is kept, the generator searches for a simpler rule that would also explain the five frames you see while pointing at a different answer. If it finds one, the question is discarded rather than shown.
- The options are also measured against each other. If two are too close to tell apart on a phone, the question is discarded: this should test reasoning, not eyesight.
Numerical reasoning
A numerical question is a table or chart plus exactly one calculation over it.
The calculation is always one of these:
- Percentage change
- Share of a total
- Row or column total
- Ratio of two cells
- Rate applied to a figure
The answer is computed from the figures in exact decimal arithmetic, never typed in by hand, so it is right by construction rather than because somebody did the sum carefully.
The answer is computed from the figures. Each wrong option is a specific mistake, built on purpose.
- Wrong options are the mistakes this test actually punishes: percentage points given where a percentage change was asked, two figures divided the wrong way round, the rate in the question ignored, the result out by a factor of ten.
- No option may land within half a percent of the correct one. A near-tie would turn a reasoning question into a rounding question.
- The wording is checked for stray digits. No number may appear in the question unless it is a label of the table itself, so nothing can be answered by matching numbers in the text.
- On chart questions every bar lands exactly on a gridline, so a value can be read rather than estimated.
Verbal reasoning
Verbal passages are written one at a time, by hand. The correct answer has to follow from one specific sentence, which is stored with the question and checked word for word against the passage you are shown.
Each wrong option is one named fault, never filler:
- Figure distorted
- Claim reversed
- Condition read backwards
- Correlation made a cause
- Necessary read as sufficient
- Beyond the passage
The correct answer is tied to one sentence of the passage, checked word for word.
- When a question needs arithmetic over figures in the passage, the working is stored and re-run. Every figure has to appear in the text and the conclusion has to hold. If the sum is wrong the whole batch fails and nothing is published.
- Nobody decides which position the correct answer sits in. The answer is named separately and the options are shuffled when the question is imported, so there is no pattern to learn.
How each question's quality is checked
These checks run over every batch, and a question that fails one is thrown away rather than patched. Far more questions are generated than are ever kept.
Most generated questions never reach practice. One that fails a check is discarded, not patched.
- The answer is derived from the same rule or calculation that produced the question, never asserted next to it.
- Every wrong option is a named, plausible mistake rather than filler.
- No option duplicates another, and no abstract answer repeats a figure already shown in its series.
- Translations are compared against the English original: same figures, same number of options, quotes still verbatim. One drifted word can change which answer is correct.
- Verbal questions are read by a person before approval. Generated numerical and abstract questions are approved by the checks themselves, then spot-read by hand.
What we do not claim
- These are not real exam questions, and nothing is copied or recalled from an EPSO test. We have no affiliation with EPSO or the EU institutions.
- Difficulty labels are estimated from a question’s structure, not from how many people got it wrong. Real answers will correct them.
- The abstract check rules out every alternative rule it can express. A series can always be read in a way the checker has no vocabulary for, so we look for that by hand.
- A good score here is evidence that you are prepared, not a prediction of your result.
Sources
The format we match comes from EPSO’s own published material. These are the primary sources we draw from.
Try it before you trust it
Unlimited practice on all three tests, free, and a full timed mock scored against both gates.