Every candidate does the same arithmetic at some point. My question bank says 71 per cent. The pass mark is around 65 per cent. Therefore I am passing. That inference is almost always wrong, and it is wrong in the same direction every time, because a question bank percentage is built from conditions that flatter you and an exam is built from conditions that do not. The number is not useless. It is simply not measuring what you think it is measuring, and knowing precisely how it deceives you is what lets you use it properly.
Key takeaways
- Your headline percentage blends first attempts with repeats, which massively inflates it.
- It reflects the topics you chose to attempt, which are the topics you are best at.
- It is usually earned untimed, with the explanation available, which the exam does not offer.
- Only first-attempt, unseen, timed, unfiltered performance approximates an exam score.
- Once official mock material is available, it supersedes your internal dashboard entirely.
The four inflators
Four separate mechanisms push your percentage above your true standard, and they compound.
Repeats. Once you have seen a question, answered it, and read the explanation, getting it right again measures recognition, not knowledge. If your dashboard blends first attempts and repeats into a single figure, and most do by default, then the more you revise the more inflated the number becomes, and the inflation accelerates as your bank fills with seen questions.
Selection. You chose which questions to attempt. Even without meaning to, candidates gravitate towards the domains they enjoy and are good at. Your percentage is therefore an average over a non-random sample, weighted towards your strengths. The exam samples from the blueprint, which is not weighted towards your strengths at all.
Timing. Most question bank practice is done untimed, or effectively so, with as long as you like to work through a stem, weigh the options, and change your mind. The exam gives you between forty-six seconds and a couple of minutes depending on the paper. Accuracy under no time pressure is not accuracy under time pressure, and the gap is often large.
Assistance. If you read the explanation before committing, or open a reference, or glance at the answer when stuck, you have not answered the question. You have studied it. Both are legitimate activities, but only one produces a number that means anything.
What to measure instead
Discard the headline figure and track five things separately. Each answers a question the percentage cannot.
First-attempt accuracy on unseen questions. This is the closest thing to a real signal, and for most candidates it is substantially lower than their headline number. If your bank does not separate it, separate it yourself: the accuracy on questions you have never seen before is the only accuracy that predicts anything.
Coverage. How much of the syllabus you have actually sampled, tracked separately from how well you did. A domain you have not attempted has no accuracy, and treating a small flattering percentage as a strength is how candidates walk into an unexamined blind spot.
Speed. Your average time per question, and whether you are within the pace the real paper demands. A candidate at 75 per cent who takes three minutes a question is not at 75 per cent in an exam that allows one.
Retention. Accuracy on a concept re-tested a fortnight after you corrected it. This is the number almost nobody tracks and the one that most reliably predicts an unpleasant surprise, because a great deal of what feels learned in week three is gone by week nine.
Confidence calibration. Whether the questions you were sure about were the ones you got right. High-confidence errors are the most dangerous items in your dataset, and they are invisible in an accuracy figure.
Genuine improvement has a signature
Because the percentage is contaminated, you need a different way to know whether you are getting better, and there is one.
Real improvement shows up as rising first-attempt accuracy on unseen questions, in domains you previously scored badly in, sustained when re-tested weeks later, under time pressure, in mixed blocks.
Each element of that sentence is doing work. Rising accuracy on seen questions is recognition. Rising accuracy in your strong domains is polish. Accuracy that evaporates on re-test is not learning. Accuracy that collapses when timed is a pacing problem masquerading as knowledge. And accuracy that only appears in filtered blocks, where you already know the topic, does not survive contact with an exam that does not tell you what you are about to be asked.
If your improvement satisfies all five conditions, it is real. If it satisfies one or two, it is an artefact of how you are practising.
Simulate the exam, not your revision
The remedy is straightforward, if unwelcome. Periodically, and increasingly as the exam approaches, generate a number under conditions that resemble the exam: unseen questions only, unfiltered across the whole syllabus, timed at the real pace, no explanations until the block is finished, no references unless the exam permits them.
That number will be lower than your dashboard figure, often significantly. It is also the only number that has ever been worth anything. Candidates find this deeply unpleasant, which is precisely why they avoid doing it, and why so many are surprised on the day.
When official material supersedes everything
One final hierarchy. Every commercial question bank, including this one, is an approximation of an exam written by someone else. When the examining body publishes its own sample questions, practice papers or official mock material, that material tells you something no third-party bank can: the exact style, the exact cognitive level, and the exact way the examiners like to construct a stem.
Treat it accordingly. Sit it under realistic timing, before you have seen the answers, and treat the result as more informative than anything your dashboard says. And preserve it: work through official material once, properly, rather than repeatedly until you have memorised it, because a memorised official paper is a calibration instrument you have destroyed.
Where iatroX fits
iatroX separates the metrics that actually predict performance rather than collapsing them into one flattering figure, so you can see first-attempt accuracy on unseen questions distinctly from your overall score, track coverage as its own dimension, and run timed mocks at the real format of your exam. The adaptive engine targets genuine weak areas rather than the topics you keep choosing, spaced repetition surfaces the retention problem before the exam does, and missed questions can be opened in the Socratic Tutor, which asks you to reason before it explains. Try it with free sample questions at iatroX. For the coverage half of the picture, see using Standard mode as a syllabus audit.
Frequently asked questions
Why is my question bank percentage higher than my mock exam score? Because your bank percentage blends repeats with first attempts, reflects topics you chose to attempt, and is usually earned untimed with explanations available. A mock removes all four of those advantages, which is why it is the more honest number.
What percentage should I be scoring before my exam? The question is not answerable from a dashboard figure, because that figure is contaminated. What matters is your first-attempt accuracy on unseen questions, in mixed timed blocks, sustained on re-test. Build a comfortable margin above the pass standard on that measure.
Should I redo questions I got wrong? Yes, but understand what the repeat measures. Getting a previously seen question right demonstrates recognition rather than knowledge. The real test is whether you can answer a different question testing the same principle, which is why re-testing a concept in a new context matters more than re-attempting the item.
How should I use official sample questions? As calibration, under realistic timing, before you look at the answers, and once rather than repeatedly. They tell you the style and cognitive level of the real exam, which no commercial bank can guarantee, but only while they remain unseen.