How the check is scored
A score you cannot inspect is not worth trusting. So here is the whole rubric: what is measured, how it is weighted, where the band lines fall, and — the part most instruments leave out — what this one does not tell you.
The five constructs
| Construct | Weight | What it measures |
|---|---|---|
| Awareness | 20% | What these systems are and are not — capabilities, and above all limits. |
| Use skill | 25% | Prompting, choosing the right tool, and fitting it into work you already do. |
| Evaluation & judgement | 25% | Spotting a confident wrong answer. Verifying output against something real. |
| Ethics, risk & compliance | 15% | Privacy, the DPDP Act, intellectual property, disclosure, and sector rules. |
| Building | 15% | Assembling a working use case — automation, tools, and agents that run without you. |
Evaluation & judgement carries the joint-heaviest weight on purpose. Testing recall of AI facts is easy and nearly worthless. Testing whether someone can tell a good output from a confident wrong one is the thing actually worth measuring, and it is the skill that stops being optional as these systems get better at sounding right.
How bands are set
Each construct is scored as a percentage of the items you answered in it. Those are combined using the weights above, renormalised over the constructs you actually answered — an unanswered construct is excluded rather than counted as zero, so leaving halfway through does not band you as though you got everything wrong.
- Band 10% and aboveGetting oriented
- Band 230% and aboveFinding your footing
- Band 350% and aboveWorking with it
- Band 470% and aboveDirecting it
- Band 585% and aboveBuilding with it
You are shown a band and, once enough people in your profession have taken it, a percentile among them — never a bare raw score. A raw number invites comparison across groups that are not comparable.
Why it is a fixed form, not adaptive
Adaptive testing needs calibrated item parameters, and calibration needs on the order of a thousand real responses per item. We do not have that yet, so we do not do it. Running an adaptive engine over uncalibrated parameters produces a number that looks statistical and means nothing — which is exactly the sort of black box this check exists to be an alternative to. Everyone sits the same form; only the order of the options moves.
What this check does not measure
- Whether you can build a working system. Forty multiple-choice and scenario items cannot tell you that; only shipping something can.
- Your ability at your actual job. It measures AI readiness, not accountancy, teaching, law or engineering.
- Anything about your employment prospects or earnings. No band predicts either, and we will not imply otherwise.
- Progress over time, from a single sitting. One result is a snapshot, not a trend.
- Anything at all if you looked the answers up. It is unproctored and infinitely retakeable — gaming it only wastes your own time.
What this is, and what it is not
This is a free, unproctored, self-administered indicator. It is not a certification, and the band is not a credential — nobody verified who sat it, and nothing you produced was examined by an assessor. A credential worth the name asserts a narrow, checkable claim backed by primary evidence and issued after proctored assessment; that is a different object, issued through a different process, and we keep the two structurally separate so that neither is mistaken for the other.
Item exposure and retirement
The item bank is several times the length of any one form, so forms rotate. Delivery is counted per item, and per-item statistics — how often it is answered correctly, and how well it separates stronger from weaker candidates — decide when an item is withdrawn. A sudden shift in a form’s pass rate is treated as a leak signal, and items are retired on that evidence rather than on opinion. Withdrawing an item takes effect immediately and does not require a release.
This check shares nothing with the Chuun entrance examination — not an item, not a bank, not a route. They are separate systems, and the free one can never leak into the selective one.
On the EU AI-literacy obligation
Article 4 of the EU AI Act has, since February 2025, required providers and deployers to take measures ensuring a sufficient level of AI literacy among staff who operate these systems on their behalf, and it carries no small-company exemption. This check can give an organisation a factual baseline of where its people currently stand. It is not a compliance audit and we do not sell it as one; meeting that obligation is the organisation’s own responsibility and involves rather more than a forty-question screener.