This workflow is for paediatric trainees using BMJ OnExamination to prepare the MRCPCH Theory and Science (TAS) written paper who want to practise adaptively — steering effort toward weak areas — without letting a rising overall score hide the small, low-volume syllabus domains. The principal limitation to name up front: BMJ OnExamination does not publish a proprietary adaptive-difficulty algorithm, so "adaptive" here means using its difficulty filters, mock tests and revision plans together with your own performance data, not trusting a black-box engine.
What BMJ OnExamination offers for TAS right now
| Item | Vendor-reported position, checked 20 July 2026 |
|---|---|
| TAS question count | "930+" questions mapped to the RCPCH curriculum, based on previous exam themes |
| Price | From £34.99 (1 month) up to £149.99 (12 months) |
| Practice modes | Select Questions (choose difficulty, recap hard items), Mock Tests (recent themes and editorial-curated), Group Learning (10-question games) |
| Personalisation | "Revision plans to turn weaknesses into strengths", feedback and peer comparison |
| Adaptive AI | No published adaptive-difficulty engine; personalisation is difficulty selection, filtering and revision plans |
| Access | App for offline study |
These figures are vendor-reported and were correct at the last check on 20 July 2026; confirm the current count and price on onexamination.com before you buy. BMJ OnExamination is an established institutional platform and a strong option for TAS, but the label "adaptive" needs care: the adaptivity you get is you reading your own data and choosing the next block, which is exactly what this workflow structures.
The TAS exam: what you are actually preparing for
TAS is 100 single best answer questions in 2 hours, one paper, one mark each, no negative marking. Delivery is computer-based at an exam centre or online. TAS is the science-and-mechanisms paper of the MRCPCH theory stage: it tests the basic scientific, physiological and pharmacological principles of clinical practice, plus evidence-based practice. In practice that means mechanism of action of drugs such as antibiotics and anti-epileptics, genetics, metabolic medicine, physiology, statistics and critical appraisal, and topics — ophthalmology, tropical medicine — that clinical experience alone will not give you. It rewards book study, not ward familiarity.
RCPCH allows the three theory exams to be sat in any order, and all three must be passed before the separate MRCPCH Clinical OSCE, which no written bank prepares you for. The official RCPCH TAS specimen is your calibration anchor, with the caveat that it sits on the previous system and includes extended matching questions no longer used — treat it as content calibration, not a format rehearsal.
Baseline week: sample before you personalise
Before you let any weakness-led feed take over, complete a small, blueprint-stratified unseen sample — for example, four to six questions drawn deliberately from every syllabus domain, timed. This gives you a domain-level map of where you actually stand. If you skip this and dive straight into "practise my weak areas", the platform's difficulty selection will simply reflect wherever you happened to start, and the low-volume domains you never sampled stay invisible.
First pass: set domain floors so the algorithm cannot hide gaps
The specific risk with any weakness-led or difficulty-selected feed is that it optimises the overall number while quietly starving low-volume domains. Your rising score can be carried entirely by the big systems while metabolic medicine, ophthalmology, genetics, statistics, ethics and palliative care go unattempted. The fix is a domain floor: a minimum number of attempted questions per syllabus domain — say 15–20 — that you commit to clearing regardless of what the feed pushes at you. Write the floors down and tick them off. The overall percentage is not allowed to rise until every floor is met.
Error taxonomy: separate the six failure modes
Adaptive practice only works if you know why you missed an item, because different errors need different responses. Tag each miss:
- Knowledge gap — you did not know the science. Action: one short source read.
- Misread stem — you missed a detail. Action: name the skipped cue.
- Premature closure — you locked onto the first plausible option. Action: write the discriminating feature.
- Guideline error — outdated answer. Action: record current source, date, jurisdiction (NICE, CKS, SIGN, RCPCH, or the SmPC via the eMC).
- Calculation error — arithmetic or units. Action: redo it, note the method.
- Time-pressure error — right idea, too slow. Action: a pacing signal, not a content one.
A feed that keeps serving "hard" items is useless if most of your misses are actually stem misreads or time pressure; the taxonomy stops you from over-treating a knowledge problem you do not have.
Review interval: not everything deserves an immediate repeat
Decide the response by error type rather than repeating every miss straight away. A knowledge gap deserves a short source read then a fresh transfer question a few days later; premature closure deserves a spaced re-test; a guideline error deserves a currency note; a time-pressure error deserves nothing but faster pacing. Immediate repeats of a just-explained item mostly train recognition of that item, which is the least useful thing to rehearse.
Mixed-block switch: objective criteria
Reduce topic-filtered practice and increase timed random blocks when: every domain floor is met; your first-attempt accuracy on unseen items has been stable across two sessions; and your misses are now spread thinly rather than clustered in one system. At that point, difficulty selection has done its job and you need exam-condition random blocks — 100 items in 2 hours, roughly 72 seconds each — to rehearse pacing and cold retrieval.
Exit criteria
Stop adding new questions when you can tick all of these, not when the bank is complete:
- Coverage floor: every syllabus domain past its minimum, none neglected by the feed.
- Stable first-attempt performance: unseen accuracy steady across sessions, not a single good day.
- Pacing: 100 items comfortably inside 2 hours.
- Retention: two-to-three-week-old items still correct.
- Official-material calibration: your RCPCH TAS specimen performance matches your bank performance.
A seven-day plan for a busy trainee
BMJ OnExamination does one job — steered first-pass practice with domain floors — and iatroX supplies unseen transfer measurement. No proprietary-algorithm claims are made for either.
- Day 1: Blueprint-stratified 40-item unseen baseline; record domain accuracy.
- Day 2: Two BMJ blocks on your two weakest domains, but only after confirming both are below floor; error-code every miss.
- Day 3: Transfer practice — one fresh iatroX item per missed principle, spaced from the original.
- Day 4: Timed random 100-item block (or two 50s); review flags only.
- Day 5: Domain-floor audit — attempt any syllabus area still under its minimum; currency-check statistics and pharmacology mechanisms.
- Day 6: RCPCH TAS specimen, timed, for content calibration.
- Day 7: Rest or light spaced review.
Decision checklist: continue, supplement, switch or stop
- Continue BMJ OnExamination while domain floors are still being filled and its items are new to you.
- Supplement with an unseen bank for honest measurement, and read primary sources where TAS mechanisms need depth a bank cannot give.
- Switch primary banks only for a measurable reason — repeated near-duplicate stems or a nameable coverage gap — never novelty or a new subscription tier.
- Stop new questions when the exit criteria are met.
Three mistakes this plan is designed to stop
First, mistaking difficulty selection for true adaptivity and assuming the platform is protecting your blueprint coverage — it is not; you set the floors. Second, letting a rising overall percentage reassure you while low-volume domains stay at zero attempts. Third, treating TAS like a clinical paper and revising from ward experience instead of the science it actually tests.
Bottom line
BMJ OnExamination is a strong, well-established TAS resource, and used deliberately its difficulty filters and revision plans make efficient adaptive practice possible. But the adaptivity is yours to drive: set domain floors, protect the low-volume syllabus areas, and refuse to let the overall score stand in for coverage. Anchor on the RCPCH syllabus and specimen, use BMJ OnExamination for steered first-pass practice, and reserve an independent unseen bank such as iatroX to confirm that the science has transferred.
FAQ
Is BMJ OnExamination enough for MRCPCH Theory and Science on its own? It can serve as a candidate's main TAS bank, but "enough on its own" understates what TAS needs: because the paper tests scientific mechanisms and evidence-based practice, some depth has to come from primary sources, and readiness has to be confirmed on unseen timed blocks. Use the bank for volume and steering, then calibrate against the RCPCH specimen and fresh items before you conclude you are ready.
Which MRCPCH Theory and Science component does BMJ OnExamination not reproduce well? A standard SBA bank is weakest at the deep mechanistic reasoning TAS prizes — deriving why a drug or physiological pathway behaves as it does, rather than recognising the fact — and at genuine critical-appraisal depth. It also does not reproduce the separate MRCPCH Clinical OSCE. Supplement the mechanism and statistics work with source reading rather than more recognition practice.
How many BMJ OnExamination questions should I complete per day for MRCPCH Theory and Science? Let your review capacity set the number, which for most working trainees is about 40–60 items a day, because a TAS miss often needs a short source read to understand the mechanism, not just the answer. Volume without that mechanistic review trains recognition and leaves the science untouched.
When should I stop using BMJ OnExamination and move to mixed mocks? Move to timed random mocks once every domain floor is met, your unseen first-attempt accuracy is stable across sessions, and your misses are no longer clustered in one system. If your overall score is climbing but some low-volume domains are still unattempted, that is a signal to audit coverage first, then switch to mixed mocks.
How should I combine BMJ OnExamination with iatroX without duplicating practice? Apply the two-Q-bank rule and give each a single job: BMJ OnExamination for steered first-pass practice, iatroX as the reserved unseen-measurement layer for fresh timed transfer questions. Keep the pools separate so no item appears in both, because the second bank only measures anything if you have not already seen its questions.
Editorial notes and references
Written by Dr Kolawole Tytler, NHS GP and founder of iatroX. Last checked 20 July 2026. BMJ OnExamination figures — the 930+ question count, pricing tiers and the revision-plan and difficulty-selection features — are vendor-reported from onexamination.com and may change; verify the current details before purchase, and note that no proprietary adaptive-difficulty algorithm is claimed. Disclosure: iatroX operates a competing UK question bank; here its role is confined to the unseen-measurement and transfer-practice jobs BMJ OnExamination does not claim, not to replacing it for TAS practice or the MRCPCH Clinical OSCE. Corrections are welcome via the feedback route on iatrox.com.
References: RCPCH, Theory exams — structure and syllabi and theory exam sample papers; BMJ OnExamination MRCPCH TAS; iatroX, Your Q-Bank Percentage Is Not Your Exam Score; iatroX, question-bank completion is not coverage; iatroX, the two-Q-bank rule; iatroX comparison hub.
Run a fresh, timed MRCPCH Theory and Science block in iatroX →