Is Cognassist validated?

Yes, and you can check it for yourself. The cognitive assessment is standardised on 26,301 UK apprentices; its two primary decision indexes meet the British Psychological Society high-stakes reliability standard (r greater than or equal to 0.8); its validity is evidenced across three strands; and it was actively developed with an independent Science Advisory Board and reviewed by PATOSS against Joint Council for Qualifications requirements. We state the honest limits alongside: two of the three validity strands are doctoral, directional and not peer-reviewed.

How accurate is Cognassist?

Accuracy has two parts, reliability and validity. On reliability, the coefficients on the two primary decision indexes meet the BPS high-stakes standard (r greater than or equal to 0.8), and nine Performance Validity Tests guard every result. On validity, the evidence runs across three strands, from construct validity confirmed by factor analysis to directional doctoral evidence that the cognitive profile predicts real outcomes. We publish the reliability coefficients rather than a single headline number.

Is the assessment peer-reviewed?

No, and we do not claim it is. The reliability and validity evidence is documented in the Technical Manual V2.0 and the validation white-paper series, and the assessment has been reviewed by PATOSS against JCQ scientific-validity requirements. The known-groups and criterion validity papers are doctoral research, independently supervised and in preparation. That is a professional review and a body of internal and doctoral evidence, not peer-reviewed publication.

What does the science actually rest on?

Two established theories: the Cattell-Horn-Carroll (CHC) model of cognitive abilities and the Baddeley-Hitch model of working memory. The nine cognitive domains map onto CHC through factor analysis, and a convergent and discriminant correlation pattern (.01 to .59) confirms each domain measures a distinct ability.

Why does Cognassist give a cognitive profile instead of a diagnosis?

Because a profile is what a tutor can act on. A diagnosis names a condition; a profile describes how a learner takes in, holds and works with information, which is the level teaching operates at, and it points directly at the adjustment to make. It also reaches learners a label never gets to, since diagnosis depends on who happened to be assessed and whether anyone pursued it. In our own criterion research, weighing 20 variables, the cognitive scores carried more predictive signal for withdrawal than the diagnosis flags did, though that is directional doctoral evidence rather than a settled finding. The profile informs professional judgement; it does not replace it.

Does it matter what device the assessment is taken on?

Yes, more than it looks. A standardised score is interpretable against the conditions the assessment was normed under, so changing those conditions changes what you are comparing against. Screen size, whether input is touch or mouse and keyboard, and how interruptible the setting is are all real differences in how a timed task is performed. Our normative sample was collected under a consistent set of administration conditions and we scope our claims to those conditions. Any vendor supporting several should be able to show a published equivalence study across them.

No. Cognassist measures the cognitive differences associated with specific learning differences to inform a professional's judgement; it is a signal, not a verdict. It does not diagnose dyslexia and does not by itself establish disability under the Equality Act 2010. Where a learner needs formal exam access arrangements, that requires a separate Level 7 specialist assessment (a JCQ Form 8).

The evidence behind the score

You can copy the screens. You cannot copy the evidence.

Anyone can copy a screen of timed tasks. What validates Cognassist is the evidence beneath it: the norm, the reliability, the validity and the independent development. Inspect it for yourself.

Book a demo See the evidence

Two questions

There are two evidence questions, and they are not the same one

Validity asks whether the measure is right. Efficacy asks whether the support that follows works. Most of this page answers the first. The second has its own page, including what we cannot yet claim.

The four pillars

How to tell a validated assessment from a screener

You cannot be the leader who bought an unvalidated screener. These are the four questions that separate a validated cognitive assessment from an unvalidated one, asked in the order a rigorous evaluator asks them. The four sections that follow answer each for Cognassist, honest limits included.

Is it standardised on a large sample that matches the population you use it on?
A score only means something against the right reference group. Ask who the assessment was normed on, how many, and whether they resemble your learners.
A large, recent, population-matched normative sample.
Do the decision-driving indexes measure consistently?
An index that gives one answer on a good day and another on a bad day cannot support a decision. Ask for the reliability coefficients, index by index.
Published coefficients meeting a recognised high-stakes standard on the indexes that drive decisions.
Is there evidence it measures what it claims, and predicts what matters?
Reliability is consistency; validity is truth. Ask what theory the tasks rest on, whether they measure distinct abilities, and whether the results predict real outcomes.
Several strands of validity evidence, each stated with its limits.
Who, independent of the vendor, stands behind it?
A vendor's own word is not evidence. Ask who developed the science, which independent experts were involved, and what external review it has passed.
Named independent experts and a documented external review.

26,301 UK apprentices

Pillar 1, the normative sample: the 2025 operational sample behind every score, standardised on the population Cognassist actually serves.

Standardised on a working-age UK apprentice population, ages 14 to 77, and refreshed for population drift.

Matched to the further education and skills population it is used on, so a score means what it should for your learners.

This is the reference sample every score is built on, not a count of everyone assessed on the platform.

It was collected under a consistent set of administration conditions. A standardised score is read against both the population and the conditions behind it, which is why we scope our claims to those conditions rather than assuming they carry across every way an assessment might be taken. Ask any vendor supporting several the same question.

Pillar 2 of 4

Reliability you can inspect, coefficient by coefficient

An assessment that drives decisions has to measure consistently. Cognassist is built to British Psychological Society (BPS) high-stakes standards and aligned to the international standards for educational and psychological testing and to the International Test Commission. We publish the reliability coefficients so you can check them, not a single summary number.

The two primary decision indexes, Language & Numbers and Executive Function / Working Memory, meet the BPS high-stakes reliability standard (r greater than or equal to 0.8).
Index internal consistency runs .78 to .93, and corrected 30-day test-retest reliability .68 to .87 (n = 112). We publish the full range, including where an index sits below the high-stakes threshold.
Nine Performance Validity Tests guard every result, flagging responses that do not reflect genuine effort before they reach a score.

Pillar 3 of 4

Three strands of validity, each with its limits stated

Validity is the harder question: does the assessment measure what it claims, and does it predict what matters? The evidence runs in three strands. Read them together, and note that we mark exactly where each one stops. The first is the strongest; the other two are directional.

Construct validity

Technical Manual V2.0

Grounded in the Cattell-Horn-Carroll (CHC) model of cognitive abilities and the Baddeley-Hitch model of working memory.
Tasks map to CHC through factor analysis, loading onto the expected verbal, memory and visual factors.
Inter-domain correlations run .01 to .59, a convergent and discriminant pattern confirming each of the nine domains measures a distinct ability.

Known-groups (clinical) validity

Doctoral, in preparation

Measures the cognitive differences associated with specific learning differences such as dyslexia, and stays specific where none is expected: it does not flag ADHD or autistic learners as a learning-difference profile.
Evidenced in doctoral research at Newcastle University, independently supervised.
The honest limit: this is directional evidence from a thesis in preparation. It is not peer-reviewed, and we do not claim it is.

Criterion validity

Doctoral, in preparation

Early cognitive flags predict apprenticeship withdrawal, and the cognitive profile predicts it better than a diagnosis label does.
Evidenced in doctoral research at Newcastle University, independently supervised.
The honest limit: a small multi-provider sample, not nationally representative, and not peer-reviewed. The effect is modest but consistent, and probably an underestimate.

Pillar 4 of 4

Independent scrutiny, on the record

A vendor's own word is not evidence. Three independent checks stand behind the assessment.

An independent Science Advisory Board
Four externally credentialed clinical and academic experts developed the assessment with our internal psychometrics team: Professor John R. Crawford of the University of Aberdeen, Dr John Welch, Dr Clive Skilbeck and Professor Chris Petkov of Newcastle University. Task content was chosen and agreed between them and our team, and that is documented in the Technical Manual rather than asserted here. Their work was the build; the assessment is maintained since by our own psychometrics team, who own the Technical Manual and refresh the normative sample.
An independent professional-association review
The Professional Association of Teachers of Students with Specific Learning Difficulties (PATOSS) has independently reviewed the assessment against the Joint Council for Qualifications (JCQ) scientific-validity requirements. That is a professional-association review against defined criteria, not a peer-reviewed study, and we describe it as exactly that.
Aligned to recognised score conventions
Results follow the JCQ and PATOSS 7th-edition conventions for standardised scores (mean 100, standard deviation 15), the recognised basis for reading a score as below average. The Speed of Working result is accepted by most awarding organisations for reasonable adjustments at qualification and end-point assessment.

Profile, not label

A cognitive profile tells a tutor what to do. A label does not.

The question a tutor faces is not what a learner has. It is what to change on Monday morning. A diagnosis names a condition; a cognitive profile describes how someone takes in, holds and works with information, which is the level teaching actually operates at. That is why understanding the scores matters more than carrying a label.

It reaches the learners a label never gets to. Diagnosis depends on who was assessed, when, and whether anyone pursued it. Measuring cognition directly finds the many learners with a real barrier and no diagnosis, including those who would never have been referred.
It points at the adjustment. Knowing a learner is below average on working memory tells you to change how instructions are given; knowing they have a diagnosis does not, on its own, tell you anything to do differently.
Our own evidence points the same way: weighing 20 variables, the cognitive scores carried more predictive signal for withdrawal than the diagnosis flags did. That is directional doctoral evidence, as set out in the criterion strand above, not a settled finding.
It stays a signal, not a verdict. The profile informs a professional's judgement; it never replaces it, and it is not a diagnosis.

See the nine cognitive domains

Honest limits

Where the assessment stops, stated plainly

Credibility comes from what we concede. Three limits, said out loud, because they are what make the evidence above believable.

It is not a diagnosis. The assessment measures the cognitive differences associated with specific learning differences to inform professional judgement. It does not detect or diagnose dyslexia, and it does not by itself establish disability under the Equality Act 2010.
Formal exam access arrangements need a separate route. A JCQ Form 8 assessment must be completed and certified by a Level 7 specialist assessor. Cognassist is a recognised tool toward Form 8, but the recommended model is inclusive profiling and everyday reasonable adjustments for all learners, with a separate Level 7 assessment where formal exam access is required.
The battery is broad, not deep. Nine domains across three indexes cover a wide spectrum of cognition, but not the depth a formal diagnosis of a specific learning difficulty requires.

FAQ

The questions a sceptic actually asks

Is Cognassist validated?
Yes, and you can check it for yourself. The cognitive assessment is standardised on 26,301 UK apprentices; its two primary decision indexes meet the British Psychological Society high-stakes reliability standard (r greater than or equal to 0.8); its validity is evidenced across three strands; and it was actively developed with an independent Science Advisory Board and reviewed by PATOSS against Joint Council for Qualifications requirements. We state the honest limits alongside: two of the three validity strands are doctoral, directional and not peer-reviewed.
How accurate is Cognassist?
Accuracy has two parts, reliability and validity. On reliability, the coefficients on the two primary decision indexes meet the BPS high-stakes standard (r greater than or equal to 0.8), and nine Performance Validity Tests guard every result. On validity, the evidence runs across three strands, from construct validity confirmed by factor analysis to directional doctoral evidence that the cognitive profile predicts real outcomes. We publish the reliability coefficients rather than a single headline number.
Is the assessment peer-reviewed?
No, and we do not claim it is. The reliability and validity evidence is documented in the Technical Manual V2.0 and the validation white-paper series, and the assessment has been reviewed by PATOSS against JCQ scientific-validity requirements. The known-groups and criterion validity papers are doctoral research, independently supervised and in preparation. That is a professional review and a body of internal and doctoral evidence, not peer-reviewed publication.
What does the science actually rest on?
Two established theories: the Cattell-Horn-Carroll (CHC) model of cognitive abilities and the Baddeley-Hitch model of working memory. The nine cognitive domains map onto CHC through factor analysis, and a convergent and discriminant correlation pattern (.01 to .59) confirms each domain measures a distinct ability.
Why does Cognassist give a cognitive profile instead of a diagnosis?
Because a profile is what a tutor can act on. A diagnosis names a condition; a profile describes how a learner takes in, holds and works with information, which is the level teaching operates at, and it points directly at the adjustment to make. It also reaches learners a label never gets to, since diagnosis depends on who happened to be assessed and whether anyone pursued it. In our own criterion research, weighing 20 variables, the cognitive scores carried more predictive signal for withdrawal than the diagnosis flags did, though that is directional doctoral evidence rather than a settled finding. The profile informs professional judgement; it does not replace it.
Does it matter what device the assessment is taken on?
Yes, more than it looks. A standardised score is interpretable against the conditions the assessment was normed under, so changing those conditions changes what you are comparing against. Screen size, whether input is touch or mouse and keyboard, and how interruptible the setting is are all real differences in how a timed task is performed. Our normative sample was collected under a consistent set of administration conditions and we scope our claims to those conditions. Any vendor supporting several should be able to show a published equivalence study across them.
Is it a diagnosis?
No. Cognassist measures the cognitive differences associated with specific learning differences to inform a professional's judgement; it is a signal, not a verdict. It does not diagnose dyslexia and does not by itself establish disability under the Equality Act 2010. Where a learner needs formal exam access arrangements, that requires a separate Level 7 specialist assessment (a JCQ Form 8).

You have seen the evidence. Now see it in practice.

You can copy the screens. You cannot copy the evidence behind the score. Book a demo to see the assessment at work, or read how providers use it.

Book a demo See customer stories

You can copy the screens. You cannot copy the evidence.

There are two evidence questions, and they are not the same one

The assessment: is the measure right?

The plan: does the support work?

How to tell a validated assessment from a screener

Is it standardised on a large sample that matches the population you use it on?

Do the decision-driving indexes measure consistently?

Is there evidence it measures what it claims, and predicts what matters?

Who, independent of the vendor, stands behind it?