How marking works, and how we check it

How your answer is marked.

• An AI model marks each answer against the four competencies in CIMA's case study blueprints: technical, business, people and leadership. It only scores a competency the task gives room for.
• It is given the task, its requirements, your pre-seen company and, where the task has one, a marking guide. It is told to judge how well you answered the question set and applied it to the company.
• You see a practice estimate on a 0–150 scale with the pass line at 80, like CIMA's scaled results, plus strengths, the points to improve and the biggest win.
• If you think a mark is wrong, use “Think this mark is wrong?” on your feedback. A person checks every query, and if the mark changes you see the new mark, the date and the reason.

AI marks are a guide to help you improve. They are not CIMA marks and not a prediction of your result.

How we check it.

Qualified markers mark a sample of real student answers on their own, and we record their mark next to the AI's. For every answer they mark, we compare the two on the same 0–150 scale:

• Average difference: how many points apart the two marks are, on average.
• Within 10 points: the share of answers where the marks are 10 points or less apart.
• Same side of 80: the share where both say pass, or both say not yet. This matters most, because it's what tells you whether you're ready.

If a reviewer has changed a mark after a query, we compare the AI's original mark, not the corrected one. We publish the figures once at least 30 answers have a human mark, and show them as they are, good or bad.

What we've found.

Study under way: figures coming

We'll publish the agreement figures here once 30 answers have been checked by a qualified human marker. A smaller sample could make the marking look better or worse than it is.

Until then, treat the AI mark as a guide to what to improve, and look at the pattern across several answers rather than any single score.