This is for candidates relying on BMJ OnExamination's dashboard to judge whether they are ready for the MRCPCH Applied Knowledge in Practice (AKP) papers. The principal limitation to hold in mind: a bank's home-screen percentage measures the questions you have already worked, under whatever selection you applied, and cannot by itself tell you how you will perform on unseen, mixed, timed items. This audit defines each metric, exposes the selection biases, and turns the numbers into quotas rather than false reassurance.
What BMJ OnExamination offers for MRCPCH AKP right now
Checked on 19 July 2026, every figure below is vendor-reported; confirm the current numbers on onexamination.com, because the product page and its meta description disagreed on question count at the time of writing.
| Feature | What OnExamination states (vendor-reported, 19 July 2026) |
|---|---|
| Coverage | MRCPCH: AKP, written by paediatricians with knowledge of the AKP paper and curriculum |
| Question volume | 330+ on the product page (a meta description cited "over 940") — verify the current count on the product page |
| Question selection | "Select Questions" by difficulty; revisit challenging items |
| Mocks | "Mock Tests" curated by the BMJ editorial team to recent exam themes |
| Social | "Group Learning" — daily leaderboard rounds of 10 questions |
| Access and app | Offline mobile app; free trial of 10 questions per day |
| Price | Roughly £29.99 (1 month) to £139.99 (12 months) |
Two honesty points up front. First, the question-count discrepancy is unresolved from the public page, so do not rely on a specific figure — verify it. Second, I did not find a claim of a proprietary adaptive-difficulty algorithm on the product page. This audit therefore does not assume one; it treats difficulty as a filter you control, and it flags that metrics such as percentile, predicted score and time-per-item may or may not be exposed on your current dashboard — check which you actually have.
The AKP exam anchor
AKP is, for most candidates, the second written stage of MRCPCH, before the Clinical (OSCE). RCPCH sets it as 120 single-best-answer questions across two papers of 60, each two hours and thirty minutes, both on the same day, computer-based, one mark per correct answer, no negative marking. It is the most applied theory paper: case scenarios, images and data-interpretation items. Crucially for an analytics audit, RCPCH publishes the syllabus content areas (roughly 23 domains) and official example questions but no fixed per-domain question count. There is therefore no official weighting for your dashboard to match — so a credible coverage audit is breadth-based, using the blueprint-coverage matrix method, not an attempt to reproduce a quota that does not exist.
Define every metric before you trust it
| Metric | What it measures | How to read it honestly |
|---|---|---|
| First-attempt accuracy | Correct on the first sighting of an item | The closest bank proxy for real performance; track it, not the blended average |
| Repeat accuracy | Correct on re-attempts of seen items | Almost always inflated by recognition; never mistake it for readiness |
| Coverage | Share of content areas and items attempted | Breadth matters more than the headline percentage; unattempted domains are your real risk |
| Difficulty | The band a question is tagged to | Useful only if you deliberately sample hard items; easy-skewed practice flatters accuracy |
| Time per item | Seconds per question | Compare against about 2.5 minutes per AKP item; fast-and-right differs from fast-and-guessing |
| Percentile | Your rank against other users | Depends on who else is in the cohort and when; a moving reference, not an absolute |
| Predicted score | A modelled pass-likelihood, if shown | Treat with caution; verify whether it is shown at all, and never present it as a guarantee |
The single most important line in that table is the distinction between first-attempt and repeat accuracy. A dashboard that shows 85% is meaningless if most of those attempts are repeats.
Selection bias: why your percentage is not comparable
Any selection you apply biases the number. If you filter to your weak topics to improve, you depress your average; if you drift toward comfortable, easy items, you inflate it. Either way, the raw home-screen percentage is not comparable with a mixed, unseen, timed block — which is the only condition that resembles the exam. Even without a proprietary adaptive engine, the act of choosing what to practise skews the statistic. That is why a readiness signal has to come from a standardised condition, not from the running average of self-selected practice.
Blueprint audit: check distribution, not just the average
Export or note your attempted-question distribution across the roughly 23 content areas and compare it with the syllabus breadth. Most candidates discover they have over-sampled a few comfortable domains and barely touched several others — commonly community child health, statistics and evidence-based practice, dermatology, ophthalmology and genetics. The home-screen average hides this completely. The fix is a coverage matrix: one row per content area, columns for attempted, first-attempt accuracy and last-reviewed date. Empty rows are your priority, regardless of what the overall percentage says.
The readiness test
A credible readiness signal for AKP requires five conditions at once: the questions are unseen (not previously worked), timed (about 2.5 minutes per item), mixed (all domains, not a single topic), unassisted (no tutor, notes or hints) and a sufficiently large sample (a full 60-question paper, not ten items). Miss any one and the signal degrades. OnExamination's mock tests can supply the timed, mixed condition; the unseen condition is the one that erodes as you work the bank, which is where a second, unseen source earns its place.
Algorithm override rules
Because difficulty and topic selection are under your control, override the comfortable defaults deliberately:
- Force the low-volume domains the bank offers few of — genetics, metabolic, ophthalmology, ENT — until each is attempted.
- Force image and data-interpretation items specifically, since AKP is image-rich and these are easy to under-practise.
- Force ethics, safeguarding and community child health, which candidates routinely skip.
- Force a hard-difficulty block weekly, so your accuracy is not propped up by easy items.
Worked dashboard example: turn analytics into quotas
Suppose your dashboard shows: first-attempt accuracy 63% over 1,200 attempts; repeat accuracy 88%; 18 of 23 content areas touched; neonatology 40 items but community child health 6, statistics 4, dermatology 3, ophthalmology 2; difficulty skewed easy/medium; average 95 seconds per item. Here is next week's plan — quotas, not predictions.
| Signal in the data | Next week's quota |
|---|---|
| Five content areas unattempted | At least 15 questions in each, this week |
| Under-sampled domains (community, statistics, derm, ophthal) | Top each up to 20+ attempted items |
| Difficulty skewed easy | One 30-item hard-difficulty block |
| Fast average time | One timed 60-question mock to test pace under pressure |
| Repeat accuracy inflating the average | 25 unseen mixed questions in iatroX, unassisted, for a clean first-attempt read |
Notice there is no predicted pass mark here. The numbers set actions; they do not forecast the exam. If you want the reasoning behind refusing to convert a bank percentage into a pass prediction, it is set out in "Your Q-Bank Percentage Is Not Your Exam Score".
Worked example: a seven-day plan around clinical work
Assume a trainee working full time, four weeks from AKP, using OnExamination as the primary bank.
| Day | Job (about 60 min) | Tool |
|---|---|---|
| Mon | Coverage: force two unattempted domains | OnExamination (Select Questions) |
| Tue | Difficulty: 30-item hard block, review every miss | OnExamination |
| Wed | Applied: image and data-interpretation items only | OnExamination |
| Thu | Transfer: 25 unseen mixed questions, unassisted, timed | iatroX |
| Fri | Simulate: 60-question timed mock, 2h30m | OnExamination mock |
| Sat | Feedback: misconception notes; update the coverage matrix | Your log |
| Sun | Retest: spaced recall of the week's misses | iatroX / OnExamination |
OnExamination does one clearly defined job — coverage and difficulty control on a large seen bank — while iatroX provides the unseen, unassisted first-attempt read on Thursday and Sunday, with no proprietary-algorithm claim. Keeping those roles separate is the two-Q-bank rule in action.
Decision checklist
- Continue with OnExamination if your coverage matrix is filling and first-attempt accuracy on unseen items is rising.
- Supplement with an unseen bank (iatroX) when repeat accuracy is high but unseen accuracy lags — the classic recognition trap.
- Switch or add a bank if whole content areas are thinly stocked, or if image and data items are scarce relative to the real paper.
- Stop drilling a domain when it is broadly covered, at first-attempt strength on unseen items and at exam pace.
Bottom line
BMJ OnExamination's AKP analytics are useful for coverage and difficulty control, but the home-screen percentage is not a readiness signal, its question count should be verified on the page, and its more advanced metrics may or may not be present on your dashboard. Audit distribution rather than the average, override the comfortable defaults, and take your readiness read from unseen, timed, mixed, unassisted blocks. Use the iatroX comparison hub if you are weighing it against another AKP bank.
FAQ
Is BMJ OnExamination enough for MRCPCH Applied Knowledge in Practice on its own? It can serve as a primary bank, but "enough" depends on coverage and on how you read the analytics. With a question count that needs verifying and no confirmed adaptive engine, the risk is a flattering average from repeats and easy items; pairing it with an unseen-measurement layer and auditing coverage matters more than the headline percentage.
Which MRCPCH Applied Knowledge in Practice component does BMJ OnExamination not reproduce well? The unseen condition, once you have worked the bank, and potentially the density of image and data-interpretation items that characterise AKP — verify how many the bank includes. Its mock tests can reproduce timing and mixing, but confirm they mirror the two-paper, 120-question structure rather than a single short set.
How many BMJ OnExamination questions should I complete per day for MRCPCH Applied Knowledge in Practice? A reviewed 40 to 60 per day around clinical work is a sustainable target, but the number matters less than what you do with it: force uncovered domains, sample hard items, and review every miss into a corrected rule. A large unreviewed count inflates the dashboard without moving first-attempt accuracy.
When should I stop using BMJ OnExamination and move to mixed mocks? Move to full, timed, mixed mocks once your coverage matrix is broadly filled and first-attempt accuracy on unseen items is stable — usually the final two to three weeks. Ring-fence at least one 60-question timed mock and one unseen mixed block for that window rather than spending them early.
How should I combine BMJ OnExamination with iatroX without duplicating practice? Assign each a single job: OnExamination for coverage and difficulty control on its bank, iatroX for unseen, unassisted, timed measurement. Because iatroX supplies questions you have not worked in OnExamination, the two measure different things rather than repeating the same items, which is exactly what prevents duplication.
Editorial notes and references
Written by Dr Kolawole Tytler, NHS GP and founder of iatroX. Last checked 19 July 2026; vendor figures (question counts, prices, access periods and features) are vendor-reported and change without notice — and at the time of writing the OnExamination product page and its meta description disagreed on the AKP question count, so verify it before relying on it. Disclosure: iatroX operates a UK clinical question bank that competes with BMJ OnExamination; it does not publish a dedicated MRCPCH bank, and its role here is confined to unseen-question measurement and transfer practice — a job the audited product does not claim. No proprietary-algorithm claims are made for either platform. Corrections are welcome via the feedback route on iatrox.com.
References:
- RCPCH, Theory examinations — structure and syllabi: https://www.rcpch.ac.uk/education-careers/examinations/theory/structure-syllabi
- BMJ OnExamination MRCPCH: AKP (product page): https://www.onexamination.com/products/mrcpch-akp
- iatroX, "Your Q-Bank Percentage Is Not Your Exam Score": https://www.iatrox.com/blog/qbank-percentage-not-your-exam-score
- iatroX, question-bank completion is not coverage: https://www.iatrox.com/blog/question-bank-completion-is-not-coverage-how-to-build-a-blueprint-coverage-matrix-for-any-medical-exam
- iatroX comparison hub: https://www.iatrox.com/compare
Run a fresh, timed MRCPCH Applied Knowledge in Practice block in iatroX →