The JobCannon assessment response corpus
774,959 completed assessments, 205 published instruments, 40 languages and 220 countries and territories — and it grows while you read this page. Everything below is an aggregate computed from the production database.
Live count
As of October 10, 2026, refreshed hourly- 774,959
- Results now
- 10,137
- Last 24 hours
- 45
- Today (UTC)
- 671,760
- Snapshot 2026-09-30
Quote the snapshot, not the live counter. The live figure is true right now and will be wrong by tomorrow; 103,199 results have been added since the 2026-09-30 snapshot alone. The dated snapshot is the number that will still reconcile when someone checks your citation.
Corpus metadata
Instruments by volume
Top 15 of 154Per-instrument score distributions — which results are common, which are rare, and how the split moves between languages — are published on the distributions pages, and internal-consistency reliability per scale on the reliability page.
Where respondents are
Top 20 of 220Read this as session geography, not respondent nationality. It is derived from the IP address of anonymous sessions that completed at least one test — 483,322 of them across 220 countries and territories — and a result row carries no country of its own. VPNs, travel and shared connections all land in here, and one person can be more than one session.
Languages taken in
Top 12 of 3164.6% of results carrying a language tag — 426,224 of 659,925 — were taken in a language other than English. That is what makes cross-language comparison on the same instrument possible at all. Results with no language recorded are excluded from both sides of that share rather than counted as English: we do not know what language they were taken in, and defaulting the unknown to the majority is how a corpus gets described as more English than it is.
How people actually take them
- 1 min 53 s
- 25th percentile
- 3 min 25 s
- Median time
- 6 min 12 s
- 75th percentile
- 83.3%
- Finish once started
Times are measured over 634,815 results that recorded a duration, not estimated from question counts. The completion rate is measured on the instrumentation event stream — 812,310 completion events against 975,489 start events — which is a different counter from stored results and does not reconcile with it row for row. It is the honest way to state “how many people who begin a test get to the end”; it is not “the share of visitors whose result we hold”.
What we publish, and what we don’t
Published on this site
- Everything on this page
- Score distributions per instrument and per language
- Internal-consistency reliability per scale
- The Big Five technical manual in full
- Two frozen anonymised response datasets — 54,431 and 77,284 rows — released under CC BY 4.0, mirrored with DOIs
Beyond the public data
- Computed results and reports only
Licence and how to cite us
Everything published on this page may be reused, including commercially, on one condition: say where it came from.
Licence
CC BY 4.0Creative Commons Attribution 4.0 on the aggregates published here — the counts, distributions, reliability figures, geography, language mix and completion times. Copy them, chart them, put them in a paper, a deck or a product. The only requirement is attribution.
The two published response datasets carry the same licence, stated on their own pages.
Ready-made attribution
Under a figure, in a chart caption or in a slide footer — this line, as it stands:
Data: JobCannon Assessment Response Corpus (snapshot 2026-09-30), https://jobcannon.io/research/data
In a reference list, the full form:
JobCannon (2026-09-30). JobCannon Assessment Response Corpus (snapshot of 671,760 completed assessment results, 2026-02-22 to 2026-09-30). Retrieved from https://jobcannon.io/research/data.
Two details that are not decoration. Name JobCannon in the running text where the figure appears, not only in the reference list — a reader should be able to see whose data they are looking at without turning to the back. And keep the snapshot date: the corpus grows every day, so a figure quoted without its date cannot be reconciled by anyone checking it later.
The licence above covers what has been published: the aggregates on this page and the two response datasets as released. It does not extend to anything we have not published. We do not sell or license individual-level data; new analyses and evaluation reports are commissioned from us. See our terms.
Common questions
How many people have taken a JobCannon assessment?
671,760 assessments had been completed as of 2026-09-30, the date of the published snapshot; the live figure on this page is higher and updates hourly. Results were collected between 2026-02-22 and 2026-09-30 across 154 of the 205 published instruments.
What is in the JobCannon response corpus?
Completed, scored results from 154 of the 205 published self-administered assessments — personality inventories, cognitive tests and interest inventories — taken in 31 interface languages by sessions from 220 countries and territories. Each result carries the instrument, the scored output, the interface language and how long it took. Names, emails and free text are not part of any published aggregate.
Is the JobCannon dataset representative of the general population?
No. It is a convenience sample of people who chose to take a test on jobcannon.io — self-selected, unproctored, unpaid and not quota-balanced on age, sex, education or country. Its value is scale, language coverage and reliability published for 20 of the 205 assessments, not representativeness. Any comparison to a population norm has to be made carefully and said out loud.
How long does a JobCannon assessment take to complete?
The median completed assessment takes 3 min 25 s, with the middle half falling between 1 min 53 s and 6 min 12 s. That is measured over 634,815 results that recorded a duration, not estimated from question counts.
Which JobCannon response datasets have been published?
Two frozen, anonymised response datasets have been published under CC BY 4.0 — 54,431 rows across the psychometric instruments and 77,284 rows of entertainment quizzes — each linked from its own page with its DOI. The live corpus itself is not published as a dataset: its aggregates are on this site.
Is JobCannon assessment data reliable?
Internal-consistency reliability is computed from our own answer vectors and published per scale for 20 assessments, last recomputed 2026-09-30 — including the scales that score badly. 23 of our 205 tests implement a published, peer-reviewed instrument; a wider 144 sit on the serious shelf, which also holds typological frameworks such as MBTI and DISC that are established but not peer-reviewed in that sense. The remainder are labelled entertainment rather than mixed in.
How do I cite JobCannon research data?
Use the prepared line: Data: JobCannon Assessment Response Corpus (snapshot 2026-09-30), https://jobcannon.io/research/data. Keep the snapshot date — the corpus grows daily, so a figure quoted without one cannot be reconciled by anyone checking it later — and name JobCannon in the running text, not only in the reference list.
What licence is JobCannon research data published under?
The aggregates published on this page — counts, distributions, reliability, geography, language mix and completion times — are CC BY 4.0 (Creative Commons Attribution 4.0), which allows reuse, including commercially, with attribution. The two published response datasets carry the same licence, as released. The licence covers what has been published and nothing else. We do not sell or license individual-level data; new analyses and evaluation reports are commissioned from us.
Need more than the aggregates?
A human baseline for model evaluation is available as computed results and reports.