Manifold Grace
RESEARCH / HCI / PSYCHOMETRIC METHOD

Human Capability Benchmarking Experiment

A structured experiment for separating three different kinds of evidence: long-run behavioural observations from human–AI interaction, published population or clinical-research benchmarks, and future task scores collected under defined conditions. It is not a diagnosis and does not treat conversational impressions as psychometric results.

METHOD FIRST

How a higher-or-lower conclusion will be earned

The experiment will not label a capability “above average” or “below average” from opinion alone. Each measurable capability needs an operational task, a suitable comparison sample and a numerical result. Where a test supplies age-adjusted standard scores, those are preferred. Where only raw population data exist, the source population, age range, mean or median, dispersion and direction of better performance are recorded before comparison.

Standard score
Common index convention: mean 100, SD 15.
Scaled score
Common subtest convention: mean 10, SD 3.
T-score
Common convention: mean 50, SD 10.
Percentile
P50 is the median; P84 is about +1 SD for a normal distribution.

Experimental classification rule: below P16 = lower performance relative to the reference; P16–P84 = broad reference range; above P84 = higher performance relative to the reference. These correspond approximately to ±1 SD only for a normal distribution. Source-provided percentile cut-offs take priority where the distribution is skewed. “Mode” is retained when a source reports it, but many continuous cognitive datasets do not publish a useful modal value; median/IQR is often more appropriate for skewed timing or error data.

How the comparison engine treats mean, median, mode and individual outcomes

Mean + SD

When the participant completed the same measure under compatible conditions and the reference source supports mean/SD interpretation, the page calculates a direction-adjusted z-score: (participant − mean) / SD. If the distribution is reasonably normal, that z-score can be converted to an estimated percentile. For measures where lower values are better, such as errors or latency, the direction is reversed before interpretation.

Median + IQR

For skewed data such as reaction times and error counts, the published median and interquartile range are shown directly. The participant can be described as above or below the median in the favourable direction, but the page does not invent a z-score or percentile from median/IQR alone.

Mode

Mode is used only where a discrete score distribution makes it informative. A multimodal distribution can have several modes; if every value is unique there is no useful mode. Mode is therefore supplementary rather than the primary benchmark for continuous cognitive performance.

Published percentiles / raw samples

If a validated source provides age-banded percentile cut-offs, the participant is located directly against those cut-offs. If matching raw participant scores are available, the page can calculate an empirical midrank percentile from the supplied sample and also display that sample's mean, median, mode and SD.

Binary or prevalence outcomes

A statistic such as “93.9% succeeded” is a prevalence figure. A participant's success or failure can be compared descriptively with that rate, but it does not by itself provide a psychometric percentile.

Progress over time

Earlier professional scores are kept as dated baselines. A later score from the same instrument can be compared descriptively; where reliability and expected practice change are available, the page can also calculate a practice-adjusted Reliable Change Index. Different instruments remain separate even when they appear to measure a similar domain.

Try the observer MCQ battery ↓ Enter a prior formal assessment ↓

AT-A-GLANCE CONTROL TABLE

Capability map

The table now shows the published reference statistic wherever one has been verified. A participant result is stored separately. If a task is not genuinely compatible with the published instrument, the page says so rather than manufacturing a percentile. Reference codes are expanded in the Sources dropdown below.

CapabilityOperational measure / referenceReference + participant scorePosition / evidence status
Memory, learning & reasoning
Working memoryNumeric span / updating. UKBUKB age 40–70: mean 6.70, SD 1.34 on its numeric-memory protocol.Ref 6.70±1.34
You: not assessed
Await compatible score
Immediate verbal recallWord-list immediate recall. ELSA-HCAP / ANDIHCAP supplies scored word-list variables; exact age-matched descriptive norm must match the administered protocol.Ref: age/protocol matched
You: not assessed
Not yet classified
Delayed verbal recallDelayed word-list recall. ELSA-HCAP / ANDIUse the same list/protocol at baseline and retest; practice effects matter.Ref: age/protocol matched
You: not assessed
Not yet classified
Visual memoryPairs/location errors. UKBUKB age 40–70: median 3 errors, IQR 3; lower is better.Ref median 3 (IQR 3)
You: not assessed
Median comparison only
Prospective memoryRemember-to-act task. UKBUKB age 40–70: 93.9% succeeded on its single prospective-memory item.Ref 93.9% success
You: not assessed
Prevalence, not percentile
Associative learningPaired-associate / concept-link task CANTAB-style domainRef: not yet located
You: not assessed
Not classified
Verbal comprehensionAge-adjusted standardized verbal index where licensed. Wechsler conventionIndex scores commonly use mean 100, SD 15.Ref 100±15
You: not assessed
Await formal/compatible score
Verbal fluencyWords in 60 seconds; exact letter/category matters. Murphy et al.COWAT-S means: 18–34 17.72±4.51; 35–49 17.10±4.78; further age bands in source.Ref age-banded
You: not assessed
Use source percentile table
Abstract / fluid reasoningNovel rule reasoning. UKB contextUKB verbal-numerical reasoning age 40–70: mean 5.99, SD 2.16 on a 13-item/2-minute protocol.Ref 5.99±2.16
You: not assessed
Only if protocol matched
Numerical reasoningTimed novel quantitative items. UKB contextContained within UKB verbal-numerical task; custom MCQ score is not the same instrument.Ref task-dependent
You: not assessed
Experimental until normed
Visual-spatial reasoningLine orientation / rotation / spatial sequence ANDIRef: not yet located
You: not assessed
Not classified
CategorisationSemantic category production and rule induction UK category normsRef: not yet located
You: not assessed
Not classified
Attention, speed & executive control
Processing speedTimed coding/search analogue; standardized or age-banded norm ANDI / ACSRef: not yet located
You: not assessed
Not classified
Simple reaction timeMilliseconds across repeated trials. UKB contextUKB matching-task median 546 ms, IQR 130; browser/device timing is not directly equivalent.Ref median 546 ms
You: not assessed
Context only
Choice reaction timeMilliseconds and errors across alternatives CANTAB-style domainRef: not yet located
You: not assessed
Not classified
Selective attentionTarget detection amid distractors; accuracy + latencyRef: not yet located
You: not assessed
Not classified
Sustained attentionLonger vigilance task; misses, false alarms and variabilityRef: not yet located
You: not assessed
Not classified
Divided attentionDual-task cost relative to single-task baselineRef: not yet located
You: not assessed
Not classified
Impulse / response inhibitionGo/no-go plus interference task. Murphy et al.Published Stroop/Hayling norms are age-banded and protocol-specific.Ref age-banded
You: not assessed
Await compatible task
Cognitive flexibilityRule-switch cost, accuracy after rule changeRef: not yet located
You: not assessed
Not classified
Task switchingSwitch-cost latency/error difference; trail-style sequencing ANDI / UKB-related trailsRef: not yet located
You: not assessed
Not classified
PlanningMinimum-move planning puzzles; completion and excess movesRef: not yet located
You: not assessed
Not classified
SequencingOrdered multi-step reconstruction; time + accuracyRef: not yet located
You: not assessed
Not classified
Organisation / structureInformation-organisation task; scoring rubric + reliability checkRef: not yet located
You: not assessed
Not classified
Planning & prioritisationConstraint-based scheduling task; objective quality score.Can be longitudinally compared within the same battery.Ref: none yet
You: not assessed
Experimental raw score
Error detection & correctionInjected-error review: detection rate, correction accuracy and time.No national norm for this custom task yet.Ref: none yet
You: not assessed
Experimental raw score
Metacognitive calibrationConfidence compared with actual accuracy; lower calibration error is better.Requires a benchmark collected with the same tasks/confidence scale.Ref: none yet
You: not assessed
Experimental raw score
Decision qualityConsistency with stated objectives under controlled trade-offsRef: not yet located
You: not assessed
Not classified
Risk / speed / value calibrationChoices under known probabilities; regret and consistency metricsRef: not yet located
You: not assessed
Not classified
Creative, social & emotional capability
Creative reframingAlternative-framing task; novelty, usefulness and number of valid reframesRef: not yet located
You: not assessed
Not classified
Divergent thinkingAlternative-uses style task; fluency, flexibility, originalityRef: not yet located
You: not assessed
Not classified
Strategy formationNovel scenario planning scored against explicit constraintsRef: not yet located
You: not assessed
Not classified
Perspective-takingMulti-agent scenario inference; competing-viewpoint accuracy.Custom observer items measure reasoning, not diagnosis.Ref: none yet
You: not assessed
Experimental raw score
Social cognitionSocial inference / intention task; accuracy and uncertaintyRef: not yet located
You: not assessed
Not classified
Emotion recognitionValidated emotion-identification task where appropriate CANTAB domainRef: not yet located
You: not assessed
Not classified
Empathic reasoningScenario-based identification of another person's likely needs; separate from self-report empathyRef: not yet located
You: not assessed
Not classified
Emotionality / reactivitySelf-report + repeated behavioural context; not inferred from “shouting” aloneRef: not yet located
You: not assessed
Not classified
Emotion regulationValidated self-report or controlled reappraisal task; distinct from emotionalityRef: not yet located
You: not assessed
Not classified
Stress tolerancePerformance stability under mild time pressure; subjective cost reported separatelyRef: not yet located
You: not assessed
Not classified
Self-regulation, work style & adaptive functioning
Self-disciplineLongitudinal behaviour + validated trait measure; not equated with overworkRef: not yet located
You: not assessed
Not classified
Distraction resistancePerformance loss with irrelevant stimuliRef: not yet located
You: not assessed
Not classified
Task initiationLatency to begin predefined tasks across interest levelsRef: not yet located
You: not assessed
Not classified
Goal maintenanceCompletion of delayed multi-step goal with interruptionsRef: not yet located
You: not assessed
Not classified
Time managementPlanned-versus-actual duration and deadline adherence across repeated tasksRef: not yet located
You: not assessed
Not classified
PersistencePersistence after failure or complexity; benefit and time-cost recorded separatelyRef: not yet located
You: not assessed
Not classified
Resilience / recoveryRecovery after setback across repeated real tasks; not a single cognitive scoreRef: not yet located
You: not assessed
Not classified
Learning agilityImprovement slope after feedback across repeated unfamiliar tasksRef: not yet located
You: not assessed
Not classified
AdaptabilityPerformance after changed rules, context or toolsRef: not yet located
You: not assessed
Not classified
CommunicationAccuracy, clarity, audience adaptation and information retentionRef: not yet located
You: not assessed
Not classified
Independent workCompletion quality under low supervision; context-sensitiveRef: not yet located
You: not assessed
Not classified
CollaborationShared-task contribution, coordination and correction; requires multi-person evidenceRef: not yet located
You: not assessed
Not classified

Loading assessment status…

Current conclusion: each score has an explicit assessment status. Published averages/medians are shown where verified; otherwise the table explicitly says that no compatible norm has yet been located. A higher/lower classification appears only after a compatible numerical result exists.

NUMERICAL COMPARISON ENGINE

How population averages and individual scores are compared

Mean + SD

When the participant completes the same or genuinely compatible task, the page calculates z = (score − mean) / SD. If lower scores are better, the direction is reversed. A normal-distribution percentile can then be estimated from z only when that assumption is supported. Mean and SD alone do not establish the shape of the distribution. Example conventions: z=0 is the mean; z≈+1 is about P84; z≈−1 is about P16.

Median + IQR

For skewed data such as reaction-time or error counts, median and IQR are retained. The page does not turn them into a z-score unless the original distribution or validated percentile table supports that step. A raw difference from the median is not automatically a percentile.

Mode & empirical percentile

The mode is the most frequent observed value; ties can yield multiple modes and all-unique data have no useful mode. It does not mean someone’s cognitive peaks. For matching participant-level scores, this page uses the midrank rule: 100 × (number below your score + half the number tied) / sample size. This describes the supplied sample, with ties made explicit.

Proportions

A reference such as “93.9% succeeded” describes prevalence. One person's pass/fail result cannot be converted into a meaningful percentile from that percentage alone.

Standard scores

Formal reports may already provide standard, scaled, T or percentile scores. Those can be placed directly on their published scale (for example 100±15 or 10±3) without reverse-engineering raw test items.

Progress from a previous formal assessment

Progress is strongest when the same standardized instrument is repeated under comparable conditions. A raw difference can be reported descriptively; where reliability and test–retest/practice-effect data are available, a Reliable Change Index (RCI) or standardized regression method can test whether change exceeds expected measurement error and practice effects. Scores from different instruments can be displayed side by side after standardization, but should not automatically be called “improvement”.

Private assessment baseline & follow-up

Earlier professional reports are baseline anchors. If you already have a psychologist's or clinician's report containing dated standard scores, scaled scores, T-scores, percentiles, raw scores or published reference statistics, enter only the fields needed for comparison below. Keep each report as a separate dated record. A future reassessment can then be compared with that baseline where the instrument and conditions are genuinely compatible.

The public page does not contain clinician names, confidential report text or private scores. Those details are entered locally by the participant if they choose. A new website quiz is a separate experimental measure and does not replace a professional assessment.

Values stay on this browser and device. Saving across visits is optional; on shared devices leave it off. The page does not submit scores or reports to a server. Browser storage is not encrypted and is accessible to scripts from this website.


Enter a dated earlier score and its published mean/SD. Leave current score empty to save only the baseline.
Which datasets can support which comparisons?

Employment, housing and education statistics describe social outcomes and barriers. They do not supply cognitive-test norms or predict an individual’s success from a quiz score. Neurodivergent is a broad category; autistic, ADHD, all-disabled and non-disabled comparison groups must retain their original definitions.

  • Buckland Review and Commons employment briefing: population employment context; retain source year, denominator, age band and definitions.
  • ONS disability outcomes (2021): inspect the specific table and impairment categories before claiming an autism or ADHD comparison. “Non-disabled” does not mean “neurotypical”.
  • ELSA-HCAP: dementia and cognitive ageing research. HCAP1 sampled adults aged 65 and over; it is not an age-matched norm for younger adults. Use documented study weights for population estimation.
  • UK Biobank Showcase: protocol-specific research data. A website MCQ or tap task cannot inherit its percentiles merely because a domain name matches.

For each dataset, preserve version, collection date, sample size, inclusion criteria, age range, measure, units, administration conditions, missing-data rule and weighting. Calculate comparisons separately for compatible cohorts; do not pool different batteries. Mean, median and mode need actual distributions or explicit published values: a data dictionary alone cannot supply them.

Clinical baselines retain their original assessment date and instrument. Follow-ups track change on compatible measures; the website quiz begins its own separate experimental timeline. Practice, measurement error and changed test conditions remain visible when interpreting change.

SPACE-SAVING TEST PLAN

Candidate tests to run

These are non-diagnostic research tasks. Where a proprietary clinical instrument is the appropriate comparator, the site will not reproduce protected test items; it will instead use an original analogue or recommend formal administration by a qualified professional if a clinical interpretation is needed.

Observer mini-battery · try it yourself

Quiz your way through the five domains, then see your own profile below. Complete all 15 items before scoring. This short untimed set is version MCQ-v1, an educational exploration with three items per domain; it cannot establish IQ, a diagnosis or a career ceiling. Repeating the same questions can produce practice effects.

Visitors can work through this short original MCQ battery and obtain raw domain scores. Completing it updates the relevant rows in the table on that visitor's browser only. Because these items do not yet have a representative UK normative sample, the result is labelled experimental rather than falsely described as a national percentile.

15 questions · five domains · four options each.

1 · Verbal: “Transparent” is to “opaque” as “explicit” is to:

2 · Verbal: Which statement preserves the distinction between evidence and interpretation?

3 · Verbal: “Allocate” most nearly means:

4 · Numerical: A task takes 24 minutes. A process improvement reduces time by 25%. New time?

5 · Numerical: Sequence: 3, 6, 12, 24, ?

6 · Numerical: If 5 records contain 2 errors each, and 3 more records contain 1 error each, total errors?

7 · Abstract: All K are M. No M are R. Which must be true?

8 · Abstract: If a rule changes from “sort by colour” to “sort by shape,” the best first action is:

9 · Abstract: Which relationship is strongest evidence of transmission?

10 · Error detection: A report says “12 participants completed the study” and later says “all 13 participants completed every task.” What should be flagged?

11 · Error detection: Which claim overreaches the evidence “A emailed B before policy C changed”?

12 · Error detection: A table labels milliseconds as seconds. This is primarily:

13 · Perspective: Two departments disagree because one optimises cost and the other safety. Best interpretation?

14 · Perspective: A manager does not mention a worker's contribution. Which conclusion is safest?

15 · Perspective: When evidence supports several explanations, the strongest response is:

Your A4 battery summary

Complete the battery to produce your own clickable, one-page profile, styled like the Digital CV. Each link is a dated snapshot of that participant’s results; it does not replace the live chart.

Formal scores are excluded by default. The generated link contains the included results; anyone you give it to can read them. No report is uploaded. Copy only the summary you wish to share.

Your summary link will appear after a completed battery.

No observer-battery result yet.

Compare with a documented sample

Start with your own result: take the battery, see your domain scores, then compare with a documented matching sample when available. There is no automatic collection of visitors’ answers or shared leaderboard.

Published averages are context, not a live connection to repository microdata. To calculate mean, median, mode and an empirical position, enter anonymised scores for one domain of this exact MCQ-v1 battery (0–3, one score per person). A convenience sample positions you within that sample only; it is not a UK, neurotypical or neurodivergent population norm. Do not paste names or confidential reports.

No sample loaded. Mean, median, mode and sample percentile are unavailable until compatible data are supplied.

Reaction-time demonstrator

This browser task produces a personal median across five trials. The UK Biobank figure is displayed as context only because its matching task, hardware and sample are different.

Not started.
Memory battery

Immediate recall, delayed recall, working-memory updating, visual-location memory and prospective memory. Record raw score, retention percentage and confidence.

Reaction & processing speed

Simple and choice reaction-time trials, using median latency, error rate and within-person variability.

Attention & inhibition

Visual search, sustained target detection, go/no-go and a colour-word interference analogue.

Switching & flexibility

Alternating-rule and number/letter sequencing tasks; measure switch cost and errors.

Verbal fluency & categorisation

Timed semantic and letter-generation tasks plus category-boundary judgements; compare only with compatible norms.

Reasoning

Original verbal, numerical, abstract and visual-spatial items with difficulty calibration across repeated sets.

Planning & prioritisation

Constraint-based scheduling and minimum-move puzzles. Score feasibility, efficiency and unnecessary steps.

Error detection

Review a controlled document or system description containing seeded inconsistencies; score detection and correction.

Metacognition

After each task, estimate confidence. Compare confidence with actual performance to calculate calibration error.

Creative association

Alternative uses, reframing and remote-association-style original items. Separate idea quantity from novelty and usefulness.

Social reasoning

Short scenarios with incomplete information; score perspective-taking, uncertainty handling and avoidance of unsupported motive claims.

Work-style longitudinal measures

Track planned versus actual duration, start latency, interruptions, completion, recovery after setbacks and learning slope across real projects.

RESEARCH CONTEXT · FUTURE EXTENSION

From task performance to brain-network research

Computational neuroscience offers a future route for studying how independently measured brain features relate to task performance. The present experiment collects an original untimed MCQ battery and a separate browser reaction-time task; it does not collect EEG, MRI or a connectome. Its chart represents observed task scores, with professional assessment values entered separately and privately.

Three measurement layers & their boundaries
LayerMeasurementsData sourceInterpretation boundary
1 · Behavioural · availableMCQ accuracy by domain and dated repeated attempts; median across five reaction-time trials. Per-item MCQ reaction timing is not currently collected.Visitor’s browser session; optional device storage.Performance on these tasks and conditions. Sleep, attention, device timing and repeated-item learning may affect results; no neural or diagnostic classification follows.
2 · Psychometric · user-enteredDated professional index scores with their instrument, edition, units and reference source; compatible follow-up change.Earlier professional reports and later independently administered assessments.Interpretation depends on instrument validity, age-appropriate norms, testing conditions, retest interval, reliability and practice effects. A custom quiz is not a substitute for the original instrument.
3 · Neurobiological · futureStructural MRI features, EEG measurements or fMRI connectivity; covariate-adjusted normative deviations where a validated reference model exists.Independent laboratory acquisition and an approved research analysis pipeline.No imaging analysis is implemented here. Brain-network identity, behavioural prediction and diagnosis are distinct questions. A statistical deviation does not establish a diagnosis or reveal a trajectory of private thoughts.
Verified studies & citation scope
  • Connectivity-method benchmarking: Liu et al. (2025), Nature Methods, evaluated 239 pairwise statistics derived from 49 measures across six families, using resting-state data from 326 unrelated healthy young HCP participants. Results varied with the chosen statistic. This supports careful method selection; it is not a population norm for the quiz. Study · DOI 10.1038/s41592-025-02704-4 ↗
  • Fingerprinting and prediction: Mantwill et al. (2022), Communications Biology, found divergent connectivity signatures for identifying participants and predicting behaviour in their analyses. Successful identification therefore cannot stand in for validated behavioural prediction. Study · DOI 10.1038/s42003-022-03185-3 ↗
  • Normative modelling: Rutherford et al. (2022), Nature Protocols, describe reference models that quantify individual variation and demonstrate the PCNtoolkit workflow. This is the normative-modelling reference; the 239-statistic study above addresses a different methodological question. Protocol · DOI 10.1038/s41596-022-00696-5 ↗
  • ENIGMA: the consortium review describes coordinated working groups and harmonised multi-site analyses, including meta-analysis and pooled individual-level analysis. ENIGMA should not be described as a single freely downloadable repository; access, participant totals and available modalities depend on the specific study. A claim about a 50,000-scan corpus-callosum pipeline needs its own exact study. Consortium review · DOI 10.1038/s41398-020-0705-1 ↗

Pending verification: the proposed 20-fold corpus-callosum processing speed-up has not been established by the supplied references. It is excluded from numerical claims until a primary study identifies the anatomy, dataset, manual comparator, hardware, timing method and accuracy checks.

Requirements before a neuroimaging extension can produce individual results

The following are proposed design requirements, not implemented capabilities:

  • Define one target measurement and research question; record acquisition modality, scanner or EEG equipment, protocol, preprocessing, atlas and feature units.
  • Select a licensed reference cohort covering the participant’s relevant age range and other justified covariates; document recruitment, sample size, missingness and scanner/site effects.
  • Acquire consented individual measurements with motion/artifact quality control. A quiz score cannot be used to fabricate a brain measurement.
  • Fit and validate the reference model using held-out data; evaluate calibration, prediction uncertainty, subgroup performance and transfer to new sites before reporting an individual centile or deviation.
  • For longitudinal analysis, preserve dated acquisitions and comparable protocols; distinguish expected age-related change, measurement error and acquisition changes from a supported individual change.
  • Use an appropriate secure research environment for imaging and identifiers. This public educational page is not a clinical imaging upload service.
Peripheral considerations: what the scores can and cannot predict

Cognitive performance can be relevant to training and job performance, but no single score determines employability, business success or the “right” way of life. A 2026 review of contemporary meta-analytic evidence reported modest correlations between general mental ability and overall job performance in newer datasets, substantially below older headline estimates. Entrepreneurship research likewise finds associations between some personality patterns and entrepreneurial status or outcomes, but individual effects are generally small and heterogeneous.

Employment outcomes also reflect opportunity, health, education, discrimination, labour-market conditions, social capital, accommodations, finance, sector choice and employer practice. UK official statistics are recorded separately from the cognitive table, with their data year, age range, outcome definition and denominator. Compare group rates using percentage-point gaps or ratios; do not pool different years, surveys or impairment definitions. “Non-disabled” does not mean “neurotypical”, and autism statistics do not describe all neurodivergent people. For example, the Buckland Review reported that around 3 in 10 working-age autistic people were in employment, compared with around 5 in 10 disabled people overall and 8 in 10 non-disabled people; that is an outcome statistic, not a cognitive-performance norm.

Accordingly, high scores may support hypotheses about particular work demands; lower scores may identify support needs, compensating strengths or environments that fit better. Neither establishes a person's worth, career ceiling or entitlement to education.

Observed interaction evidence: hypothesis layer only

Longitudinal conversations can provide ecological evidence about behaviours such as sustained research, error correction, conceptual association, planning, reframing and persistence. They are valuable because they occur across real tasks, but they are not standardized conditions. This layer will therefore be stored separately from numerical psychometric results and will never be used to manufacture a percentile.

Age, education, language and test conditions

Norms can change materially with age, education, language, device, sleep, fatigue, practice effects and testing environment. A valid comparison must use the correct reference group and record the conditions under which the task was completed. Adult-only norms will not be used to classify a younger participant, and older-adult cohort data will not be treated as a universal population average.

Sources & benchmark notes
  • UKB — UK Biobank cognitive data. UK Biobank recruited adults aged 40–69 and notes healthier/wealthier volunteer selection; it is not automatically a representative UK or neurotypical reference. Cohort details ↗ A published analysis of 480,416 participants aged roughly 40–70 reported numeric-memory mean 6.70 (SD 1.34), verbal-numerical-reasoning mean 5.99 (SD 2.16), reaction-time median 546 ms (IQR 130), visual-memory errors median 3 (IQR 3), and 93.9% success on the prospective-memory item. Open paper ↗
  • Murphy — adult lifespan executive-function norms. UK neuropsychology researchers published age-banded Stroop, letter-fluency and Hayling norms across ages 18–92. For example, the 35–49 group had mean Stroop correct 119.25 (SD 20.71) and COWAT-S mean 17.10 (SD 4.78); the paper provides percentile tables for each age band. Open paper ↗
  • ELSA-HCAP / UK Data Service. The first ELSA-HCAP wave sampled 1,274 adults aged 65 and over, including selected cognition groups with study weights. It is not a general adult or younger-adult normative sample. Study 8502 provides harmonised cognitive-assessment datasets and score dictionaries covering tasks such as letter cancellation, backwards count and immediate/delayed memory measures. Some instruments are licensed; this page therefore cites the protocol rather than copying protected items. ELSA-HCAP ↗ · UKDS study 8502 ↗
  • Reliable change. Neuropsychological change research recommends comparing retest differences with measurement error and, where relevant, practice effects rather than treating every numerical increase as true improvement. Methods review ↗
  • ANDI — Advanced Neuropsychological Diagnostics Infrastructure. Aggregated control datasets include Trail Making A/B, letter and semantic fluency, Stroop, delayed verbal and visual recall, coding and other neuropsychological measures across large pooled samples. Open database paper ↗
  • UK category-production norms. UK researchers published semantic-category production data for 117 concrete and abstract categories, useful for categorisation and semantic-fluency research. Open paper ↗
  • CANTAB normative work. Cambridge Cognition reports adult normative samples of around 750 participants per collection block across episodic memory, emotion recognition, executive functioning, working memory and other domains, with age-related effects. Summary ↗
  • ACS / Wechsler score conventions. Pearson's Advanced Clinical Solutions documentation describes scaled scores with mean 10/SD 3, index scores with mean 100/SD 15, and T-scores with mean 50/SD 10. Manual chapter ↗
  • ONS — socioeconomic outcomes, not psychometric norms. ONS 2021 disability outcomes reported employment 53.5% for disabled versus 81.6% for non-disabled people, with autism at 29.0% among respondents reporting it as their main impairment; degree attainment was 24.9% versus 42.7%. ONS release ↗
  • Buckland Review of Autism Employment. The 2024 government-backed review reports around 3 in 10 working-age autistic people in employment versus about 5 in 10 disabled people overall and 8 in 10 non-disabled people. GOV.UK report ↗
  • Job-performance interpretation. A 2026 review of newer and older meta-analytic evidence reports that general mental ability predicts job performance, but contemporary corrected estimates are modest and vary with job characteristics. Journal article ↗
  • Entrepreneurship interpretation. Zhao & Seibert's meta-analysis found small individual Big Five differences between entrepreneurs and managers, with a moderate multivariate relationship across the set of traits (R=.37). PubMed record ↗