A pass mark is the minimum score a candidate must reach to pass a test or exam. It turns a score into a decision: certified or not yet. Set well, it reflects the level of knowledge a role actually needs. Set badly, it is an arbitrary line that passes people who cannot do the job or fails people who can.
Why does the pass mark matter?
For roles that oversee AI agents, a certificate may decide who can approve payments, account changes or customer decisions. The pass mark is what that certificate rests on. Too low, and approval rights go to people who have only half-learned how a governed agent decides. Too high, and the organisation cannot staff its review queue.
The mark also shapes behaviour. If people know they can miss a quarter of the questions, some will skip whole topics. A pass mark alone cannot prevent that; it needs to sit alongside other checks.
How is a pass mark set in practice?
Good pass marks are set from the role, not by habit:
- Define what a minimally competent person in the role must know.
- Write questions that test those points, and tag each to a topic.
- Ask experienced practitioners which questions a minimally competent person would get right.
- Set the mark at that level, then check it against trial results.
- Add must-pass items for knowledge that cannot be traded off, so a strong score elsewhere cannot hide a gap.
The last step matters for agent oversight. Knowing that an external document cannot authorise a payment is not something to average away.
What is an example?
A written exam for people who review an agent's work has 40 questions and a pass mark of 32 correct. A candidate scores 34 but gets both questions on held payments wrong. Under a simple pass mark, they pass. If the exam also requires a practicum on real decisions with no missed red line, the gap shows up there. That is why a pass mark is rarely the only test for a role with approval rights.
How is a pass mark different from an agent's pass floor?
A pass mark applies to a person's exam score. A pass floor applies to an agent's certification: the agent is replayed against real cases from the customer's own operation and each answer is scored against a floor, gate by gate. Every gate must clear on its own; a strong grounding score cannot make up for a weak refusal score. The design questions are set out in what score an AI agent should need to pass, how to set the pass floor for each certification gate and the pass floor setting template.
The two meet in practice. A reviewer who passed their exam approves the work of an agent that passed its floors. Both thresholds have to mean something.
Which terms are related?
- Cut score: another name for a pass mark.
- Capstone project: a pass judged on a built deliverable rather than a score; see what a capstone project is.
- Question bank: the pool an exam's questions are drawn from.
For the wider topic, read AI literacy and certification for people who work with agents.
Where the Academy fits
The Certified Pryme Operator written exam in Pryme Intelligence Academy has 40 questions in 60 minutes, with a pass mark of 32 correct. It is paired with a two-week practicum of real review decisions with no missed red line, so the exam score is never the only evidence.
Questions people ask
Should the pass mark ever be lowered to pass more people?
No. If too many fail, look at the training or the questions. Lowering the mark changes what the certificate means for everyone who holds it.
Is a higher pass mark always safer?
Not if the questions are poor. A high mark on questions that test recall rather than judgement adds little. Fix the questions first.