Because a written test cannot show that someone reviews well when it counts. Review happens at volume, with interruptions, messy records and deadlines. A candidate can score well in a quiet exam room and still approve a bad payment on the eightieth item of a busy afternoon. Only assessed real decisions show which reviewer they are.
What does a written exam miss?
A written exam is good at three things: checking that the candidate knows the concepts, testing judgement on prepared scenarios, and doing both fairly and repeatably. It is poor at four others:
- Volume. Real queues bring dozens or hundreds of items. Attention fades, and so does scrutiny.
- Noise. Live records have missing fields, odd formats and irrelevant detail. Exam questions are cleaner than reality.
- Base rates. Most held items are fine. The skill is catching the rare bad one among many good ones, which a balanced exam cannot reproduce.
- Consequence. In a live queue, a decision has an effect. People behave differently when it does.
These gaps are where incidents happen. A reviewer who approves quickly because the last 50 items were fine is the failure mode the hold exists to prevent.
What should the practical part look like?
| Element | Why it matters |
|---|---|
| Real decisions in a live or production-like workspace | Tests the actual skill, with real noise |
| A long enough period, such as two weeks | Covers ordinary days, busy days and unusual cases |
| Accuracy measured against an assessor's view | Separates good judgement from luck |
| Zero tolerance for a missed red line | Mirrors the cost of the error in live work |
| Reasons reviewed, not just outcomes | A correct choice for a wrong reason is a near miss |
| Sign-off by someone independent of the candidate's manager | Removes pressure to pass a colleague |
The independence point is easy to skip. A manager short of approvers has every reason to pass their own team. An assessor from outside the line does not.
An example: two candidates, same exam score
Take two fraud analysts who both pass the written exam comfortably. In the two-week practical, the first works steadily, declines three held card releases with clear reasons, and asks for help on one ambiguous case. The second clears the queue faster than anyone, but on day nine approves a held release where the agent had flagged a device change and a new delivery address, a pattern the team treats as a red line.
On paper they were equal. In practice, one is ready and one is not. Without the practical part, both would have been granted approval rights on the same day.
Does the practical part make certification too heavy?
It makes it slower, and that is the point. A certificate that can tie to approval rights should be hard to earn. Two things keep the load reasonable. First, the practical uses work the candidate would be doing anyway, reviewed by an assessor, so it adds assessment rather than extra work. Second, renewal does not repeat it; an update module and a short check are enough once competence has been shown, provided certificates expire, as why certificates for AI agent skills should expire explains.
For employers, the practical part is what gives the certificate value. It is why AI certificates can be worth it for employers: the employer can grant authority on the strength of it. A business can then tie approval rights to certificates in the AI agent workspace, depth by depth, knowing the certificate reflects real work. Status should be visible to anyone through a public register of certificates. The guide to AI literacy and certification for people who work with agents places the practical part in the full structure.
Where the Academy fits
The Certified Pryme Operator assessment in Pryme Intelligence Academy pairs a written exam with a two-week practicum of real review decisions. The candidate must get 19 decisions in every 20 right, with no missed red line, and a Pryme assessor signs off. Builders and Architects also finish with practical work: a capstone agent certified in Attest, or a capstone and a board review.
Questions people ask
Can a simulation replace real decisions?
A good simulation helps with practice and early assessment. It cannot fully reproduce consequence and base rates, so it works best as preparation for an assessed period of real work.
What if the candidate's queue is quiet during the practical?
Extend the period until enough decisions have been made to judge. A practical with too few decisions measures little.
Who should act as assessor?
Someone who is certified at the same level or above, is not the candidate's line manager, and reviews the reasons as well as the outcomes.