Culture
Knowledge that only one country or culture would have, such as local sports, holidays, brands or school systems. A question should test the skill, not where someone grew up.
How we build and check our tests
Every question is proved before a candidate sees it. Candidates are compared with each other and with everyone who has taken the test. Each kind of test is backed by the largest published research on what predicts job performance.
Get started free
This page explains how that works, for HR leads, hiring managers and psychologists.
WHAT WE BELIEVE
Our approach
Six beliefs shape every test we publish and every part of the platform.
01
Questions are proved before anyone sees them. Every question passes all of its checks before it reaches a candidate. We never try out new questions on real candidates, and every question a candidate answers counts towards their score.
02
Tests are built to frameworks, not by individuals. Each test follows a written plan and recognised standards, so its quality does not depend on who wrote it. Our core tests were developed and validated by occupational psychologists and industry experts, and every newer test is built to the same frameworks.
03
Candidates are compared with each other. The assessment average shows how a candidate compares with the others who applied for the same job. The global average shows how they compare with everyone who has taken the test.
04
Backed by the largest research. We choose which kinds of test we offer using the biggest published studies of what predicts job performance.
05
Fair by design. Every question is checked for culture, language, stereotypes and access as it is written. We ask candidates nothing about themselves.
06
A score is one piece of evidence. We set no pass marks. Use tests alongside your interviews.
THE LIBRARY
Our tests
Our library is organised into seven types, the same names you see in the app. Each type measures something different, and each test's own guide names and defines the skills it measures.
| Type | What it measures |
|---|---|
| Cognitive ability | Reasoning ability: numerical, verbal, deductive, abstract, mechanical and spatial reasoning, and error checking. Timed, with one correct answer per question |
| Personality | How someone prefers to work and what motivates them, including our Personality and Drives questionnaires and type profiles such as DISC. Untimed, with no right answers |
| Soft skills | Judgement in realistic work situations: communication, teamwork, problem solving, emotional intelligence and similar |
| Role specific | The knowledge a particular job needs, such as bookkeeping, customer service, sales, project management or IT support |
| Software | Working knowledge of the tools people use every day: spreadsheets, design, analytics, advertising and CRM |
| Coding | Languages, frameworks and developer tools |
| Language | Proficiency in a language, plus English grammar and spelling |
| Typing | Typing speed and accuracy, and number pad data entry. Listed under Role specific in the library |
| Case studies | Work samples: the candidate works through a realistic pack, such as a budget or a sales report, and answers questions on it. Listed under Role specific in the library |
Tests that exist at more than one level are labelled Foundation, Intermediate or Advanced. The label describes how much the task asks of a candidate: how many steps, how much has to be held in mind at once and how much structure is given.
BUILT AND CHECKED
How we build a test
Each kind of test has its own build process, suited to what it measures. They share one rule: a question that fails any check is thrown away, or rewritten and checked again from the start. Nothing goes live with a known fault.
Every new scored question goes through the same core steps:
- 01 / PLAN
A written plan comes first.
Before any question is written, the test has a specification: the skills it measures, the format, the timing and a set mix of easy, medium and hard questions, weighted towards medium so the test separates candidates across the range.
- 02 / ONE ANSWER
One defensible answer.
Every question is checked to have exactly one correct answer.
- 03 / CHECK WITHOUT THE KEY
An independent check without the answer key.
The question is answered by solvers or reviewers who do not see the key. Where they settle on a different answer from the key, the question is fixed or dropped.
- 04 / FAIRNESS REVIEW
A fairness review.
Every question is reviewed for the causes of unfairness described in the next section.
- 05 / LIVE
Only then does it go live.
Every question passes all of its checks before it reaches a candidate.
Core library
Our core tests, including most cognitive ability tests, Personality, Drives, soft skills and language tests, were developed and validated by occupational psychologists and industry experts.
Cognitive ability tests
Our newer numerical, verbal, deductive and error checking tests are built at Foundation and Advanced levels.
- Numerical questions are worked backwards from the arithmetic. The answer is fixed first and the data is built to produce it, so every figure in a table or chart is consistent.
- Wrong answers come from real mistakes. Each wrong option is the result of a named, common error, such as using the wrong base for a percentage change, so the question rewards reasoning rather than guessing.
- Formats candidates will recognise. Question styles follow the conventions of the established graduate and professional reasoning tests.
- Several independent solvers. Each question is solved separately by several solvers working without the key. If they agree with each other on a different answer, the question is fixed. If most of them miss the key without agreeing, the question is reviewed by hand before it can go live.
- Automatic giveaway checks. Every question is screened for patterns that would let a candidate find the answer without reasoning, for example a correct option that is longer or more precise than the others, or wording that repeats the question.
- Reviews that are tested themselves. Questions with known faults are mixed into each review panel, so we can confirm the review catches what it should before its verdicts are trusted.
Case studies
- Realistic work packs. Each case study gives the candidate a role brief, emails, tables and charts, as they would meet them at work.
- Every number has its formula. Every total, percentage and chart value in the pack is recalculated by script from its source figures, so the pack is internally consistent.
- An independent solve and two separate reviews, for accuracy and for craft.
- Only the strongest packs go live. More packs are written than are published, and the weaker ones are dropped after review.
Job skills, software and language tests
- Each test is written against its field's own standard, such as IFRS for accounting, ISO 45001 for health and safety, the Scrum Guide for agile delivery or GDPR for data protection. The full list is under standards we follow.
- Large pools. Each test is built with a pool of three to four times the questions a candidate sees, so candidates on the same test do not all see the same questions.
- Every wrong answer earns its place. Each wrong option has a written reason: it is either true of something else or a common mistake. Options that could be ruled out on wording alone are rewritten.
- Format and balance checks. The position of the correct answer is balanced across a test, and option lengths are checked so length gives nothing away.
- Wrong keys are planted on purpose. The independent solve includes questions with a deliberately wrong key, so we know it is reading each question rather than agreeing with the key.
- Two separate reviews, one for accuracy and one for craft, so one kind of fault does not hide another.
- Wording that works internationally, and every translated test is checked again in its new language.
Role questions from your job advert
When you build an assessment from a job advert, you can add questions written for that role.
- Set in the role's real situations, taken from the advert you provide.
- Difficulty is anchored to the library. Library questions at the same level guide how hard each question is, aiming for a question that a qualified person usually gets right and an unqualified person usually does not.
- An independent review drops any question the reviewer answers differently, any it judges too easy, any that could be answered without knowing the role, and any that fails the fairness review.
Profiles and typing
- Type profiles (16 Types, DISC and Enneagram) use plain-English statements built on the published models, with their scoring checked before release.
- Typing tests use original passages, with speed and accuracy marked on our server rather than in the browser. Pasting is blocked, and impossible speeds or machine-like rhythm are flagged.
FAIR BY DESIGN
Fairness built into every question
Fairness is checked when each question is written, not afterwards on candidates. Every new question is reviewed for four things, and a question that fails is thrown away.
Language
Idioms, slang and needlessly hard words. Outside language tests, a question should test the skill, not how good someone's English is, so wording is kept as plain as the content allows.
Stereotypes
Names, roles and situations spread across genders and backgrounds, with no group shown in a fixed role, such as every manager a man or every assistant a woman.
Access
Charts and images that can be read without relying on colour alone.
We ask candidates nothing about their age, gender, ethnicity or background. Fairness comes from how the questions are built, and from how you use the results: see using results well.
COMPARISON
The two averages
Every scored library test shows a candidate's result against two averages.
The assessment average
Compares a candidate with the others who applied for the same job. This is the comparison that matters most, because it is the choice you are making: which of these applicants to take forward.
The global average
Compares a candidate with everyone who has taken the test with us. It shows whether a whole group of applicants is stronger or weaker than usual.
Both averages are kept clean. Only a candidate's first sitting of a test counts, so a second attempt an employer allows does not count twice. Demonstration, test and cancelled sittings are left out, and so are sittings taken with extra time or without a time limit, so every result in the average was taken under the same conditions.
A new test starts from the middling score it was built to produce, through its easy, medium and hard mix. The real global average takes over as soon as enough people have taken it.
Personality results are placed against comparison bands set and validated by the occupational psychologists who built the test.
THE EVIDENCE
What research says
The largest review of what predicts job performance was published by Sackett, Zhang, Berry and Lievens in 2022. It re-examined decades of studies and corrected how earlier reviews had adjusted their figures. It ranks job knowledge tests and work samples above general cognitive ability tests.
| Our group | Kind of test in the research | 2022 figure | 1998 figure |
|---|---|---|---|
| Role specific, software, coding, language | Job knowledge test | .40 | .48 |
| Case studies, typing | Work sample | .33 | .54 |
| Cognitive ability | Cognitive ability test | .31 | .51 |
| Soft skills | Situational judgement test | .26 | not measured |
| Personality | Conscientiousness, work-focused questions | .25 | not measured |
| Emotional intelligence | Ability-based emotional intelligence | .22 | not measured |
| Personality | Conscientiousness, general questions | .19 | .31 |
| Drives and type profiles | No direct figure | none | none |
| For comparison: a CV | Years of job experience | .07 | .18 |
Each figure is a correlation between a kind of assessment and later job performance, where 1.0 would be a perfect match and 0 would be no link at all. The figures apply to each kind of test, not to any one company's test, including ours. They show which kinds of test are worth using.
The 1998 column is the earlier review by Schmidt and Hunter, which many websites still quote. The 2022 figures are lower because the newer review corrects for range restriction more carefully. The order matters more than the exact numbers: knowledge tests and work samples lead. Every kind of test with a figure in the table predicts job performance better than years of experience on a CV.
Drives and type profiles have no direct figure because they describe motivation and preferences rather than predicting performance. Use them to shape interview questions, not to rank candidates.
GOOD PRACTICE
Using results well
We set no pass marks. The right score for a role depends on the role, the applicants and what else you know about them, so that decision is yours. If you want a minimum score, set it fairly:
- Decide it from the role, before you see any scores. Ask what level of the skill the job genuinely needs on day one.
- Write it down, with the reason.
- Use it for everyone applying for that role, with no exceptions after the event.
- Check it against the assessment average. If almost every applicant falls below your bar, the bar or the test may be wrong for the role.
Use tests alongside your interviews. Each test's guide in the app suggests an interview question and what to listen for.
Treat a score as one piece of evidence. Use the same assessment for every candidate for a role, record your decision and why, and never let a score make the whole decision.
More guidance is in our help centre: how our tests are written.
ACCESS
Adjustments
Every candidate should be able to show what they can do. Employers can:
- Give everyone extra time: 25%, 50% or 100% on every timed test in an assessment.
- Give one candidate their own extra time, set against their email address, when they are invited or later.
- Ask us to switch off time limits for your account where speed does not matter for the role.
- Tell candidates before they start that they can ask for an adjustment. We recommend turning this on: it is the only way most candidates will know.
Candidates can also turn on single-key answering, which lets them work without a mouse. Personality and Drives questionnaires are untimed.
How candidates ask. A candidate asks the employer who invited them, using the contact details in their invitation email. If they would rather not explain their circumstances to the employer, they can get in touch with us and we will help them work out what to ask for.
Any adjustment is recorded with the candidate's results, so everyone reviewing them knows the conditions the test was taken under. Sittings with extra time are left out of the averages, so they do not shift the comparison for anyone else.
Full detail: timing, extra time and adjustments.
LIMITS
What we don't claim
- A score does not, on its own, predict how a person will perform in your company. It is strong evidence of one thing, measured consistently. The job, the team and the person all matter too.
- The research figures describe kinds of test, not our tests. We use them to choose which kinds of test we offer, never as a figure for any one of our tests.
- No test is the whole decision. We do not tell you who to hire, and we do not set pass marks.
- Fairness checks remove known causes of unfairness from our questions. How results are used is still your decision, and it matters as much.
STANDARDS
Standards we follow
| Standard | Where we use it |
|---|---|
| Standards for Educational and Psychological Testing (AERA, APA and NCME) | Test design, specification and fairness for every new test |
| Multiple-choice item-writing guidelines (Haladyna, Downing and Rodriguez) | How every multiple-choice question and its wrong options are written |
| Web Content Accessibility Guidelines: use of colour | Charts and images that do not rely on colour alone |
| IFRS Accounting Standards | Accounting and finance tests and finance case studies |
| ISO 45001 | Health and safety tests |
| The Scrum Guide | Agile and Scrum tests |
| ISTQB | Software testing tests |
| GDPR | Data protection content |
| DigComp, the European Digital Competence Framework | Digital skills tests |
| Kotter's 8-step model | Change management tests |
| Prosci ADKAR | Change management tests |
| Open Extended Jungian Type Scales | 16 Types profile |
| The DISC model | DISC profile |
| The Enneagram | Enneagram profile |
COMMON QUESTIONS
Common questions
Who writes your tests?
Our core tests were developed and validated by occupational psychologists and industry experts. Every newer test is built to written frameworks and the standards above, and passes the same checks before release, so quality does not depend on any one author.
Does every question count?
Yes. Every question is proved before a candidate sees it, so nothing is tried out on candidates and every answer counts towards their score.
How do you know your tests work?
Each kind of test we offer is backed by the largest published research on what predicts job performance, set out in what research says. And every question is proved before a candidate sees it, as set out in how we build a test.
Who is in the global average?
Everyone who has taken the test with us, counting only first sittings taken under standard conditions. Demonstration, test, cancelled and adjusted sittings are left out.
Why do you ask candidates nothing about themselves?
Fairness is built into each question as it is written, so we do not need personal details to check it. Candidates spend their time on the test, not on a form.
What score should I look for?
The one you set from the role. Decide it before you see any scores, and use it for everyone. See using results well.
How do you know every question has one right answer?
Our core tests were developed and validated by occupational psychologists and industry experts. Every new scored question is checked to have exactly one correct answer, then answered by solvers or reviewers who do not see the key. Where they settle on a different answer from the key, the question is fixed or dropped before any candidate sees it.
Can I see what candidates will face?
Yes. Every scored test's guide includes a sample question you can answer yourself.