Psychometric research on a live corpus of 594,590 completed assessments
594,590 completed assessments, 28 languages, 217 countries and territories — and growing every hour. We publish the aggregates openly: distributions, reliability, geography, completion times. Nobody has to take our word for a number.
- 594,590
- Completed results
- 194
- Instruments
- 28
- Languages
- 217
- Countries & territories
Snapshot taken 2026-09-22, covering 2026-02-22 to 2026-09-22. Every figure on this page is an aggregate computed directly from the production database — none of it is estimated, projected or rounded up from a smaller sample. The corpus passport shows the live count beside this snapshot, so you can see exactly how far it has moved.
What we publish
Response corpus passport
What the corpus contains and who is in it: 594,590 completed results across 194 instruments, 28 languages and 217 countries and territories, with completion times and the live counter.
Updated 2026-09-22 · live counts
Score distributions
How results actually fall per instrument — which types are common, which are rare, and how the split changes by language. Real norms, not textbook percentages.
Norms pulled 2026-08-16
Reliability (Cronbach’s alpha)
Internal-consistency reliability per scale for 23 assessments, computed from our own answer vectors. Published including the scales that score badly.
Recomputed 2026-08-06
Big Five technical manual
Depth on the one instrument procurement asks for by name: factor structure, reliability, norms, convergent evidence and the stated limitations.
Validity study
Open dataset — entertainment quizzes
77,284 anonymised responses to our for-fun quizzes in 24 languages. Free download, CC BY 4.0, mirrored with a DOI.
Download · CC BY 4.0
Hiring & AI statistics library
70 statistics on AI in hiring, screening bias and the skill economy, each one carrying a link to the paper or survey it came from.
Other people’s research
What is open, and what is not
Open to everyone
- Corpus scale, coverage window and growth
- Score distributions per instrument, and per language
- Internal-consistency reliability per scale
- Country and language mix
- Completion times and completion rates
- Two frozen anonymised response datasets — 54,431 and 77,284 rows — under CC BY 4.0
Everything in this column is CC BY 4.0: reuse it, including commercially, as long as you say where it came from. The attribution line is written for you.
Under agreement only
- Raw responses and any per-row data
- Per-item statistics, including item difficulty
- The question banks themselves
- De-identified extracts cut to a study design
- Fielding new instruments to our respondent stream
Per-item difficulty is a map of the question bank. Publishing it openly would damage the instruments it describes, so it goes to partners and nowhere else.
Working with the data
Research partnerships
Universities and labs run studies on our respondent stream: field an instrument, get de-identified data cut to your design, co-author the paper.
Human baselines for AI evaluation
Model scores only mean something against a human distribution. Ours spans 594,590 results across 217 countries and territories on instruments with published reliability.
Grant-funded work
We hold the panel, the instruments and the infrastructure; a university partner holds the research lead. That combination is what funders ask for.
Evidence for procurement
Buyers ask for reliability, norms and sample provenance before an assessment goes near a hiring decision. Every one of those is already public here.
Common questions
How large is the JobCannon assessment dataset?
As of 2026-09-22, 594,590 completed assessment results across 194 instruments, collected between 2026-02-22 and 2026-09-22. Sessions came from 217 countries and territories, and results were taken in 28 interface languages. The live count is published on the corpus page and is always higher than this snapshot.
Is JobCannon research data publicly available?
The aggregates are: scale, coverage, score distributions per instrument, internal-consistency reliability per scale, country and language mix, and completion times are all published openly on this site. Raw responses and per-row data are not public. Two frozen, anonymised response datasets are available for direct download under CC BY 4.0.
Where does the data come from?
From people taking JobCannon assessments on jobcannon.io of their own accord — self-administered, unproctored, unpaid, and not recruited into a panel. That is a convenience sample, not a representative one, and every aggregate we publish should be read as describing our respondents rather than a general population.
Can I cite JobCannon research in a paper?
Yes. The published aggregates are CC BY 4.0 — free to reuse, including commercially, with attribution — and the corpus page prints a ready-made attribution line you can paste under a figure. Cite the pull date shown beside every figure: it is there precisely so a citation still reconciles months later, because the live counter moves daily. Anything beyond the published aggregates is arranged directly with us.
How many languages and countries does the data cover?
Results have been recorded in 28 interface languages out of 28 the platform supports, and sessions completing at least one test came from 217 countries and territories. 62.8% of results carrying a language tag were taken in a language other than English, which is what makes cross-language comparison on the same instrument possible.
Can researchers get access to the underlying data?
Yes, under agreement. Beyond the public aggregates we work with universities and labs on de-identified extracts, item-level statistics, and fielding new instruments to our respondent stream. Per-item difficulty and the question banks are never published openly and are released only under a signed agreement.
Want to work with this data?
We field instruments for research partners, supply human baselines to AI labs, and work with universities on grant applications. 195 instruments, a respondent stream that never stops, and aggregates you can check before you talk to us.
How the instruments are built and validated: methodology.