This audit is for paediatric trainees who have bought — or are weighing — Pastest for the MRCPCH Applied Knowledge in Practice (AKP) paper and want to know how far its AI "Tutor mode" can be trusted. Pastest is a strong, blueprint-aware bank with a large AKP-specific question pool. Its principal limitation is not the questions: it is that the tutor's grounding is invisible, so an articulate explanation is not the same as a verified one.
What Pastest offers for MRCPCH Applied Knowledge in Practice right now
Currency first, because counts, prices and feature names move. Every figure below is vendor-reported from the Pastest AKP product page and was accurate on 19 July 2026; confirm it before you pay.
| Item | Current state (vendor-reported, checked 19 July 2026) |
|---|---|
| AKP coverage | Yes — a dedicated MRCPCH AKP bank (separate products cover FOP, TAS and Clinical) |
| Question count | Over 2,000 exam-style single-best-answer questions (vendor-reported) |
| Mock/past papers | 3 bespoke past papers (vendor-reported) |
| Learning content | 300+ textbook topics and around 50 multimedia resources (vendor-reported) |
| AI feature | "Tutor mode" — instant, on-demand AI support while you answer, opened from a side panel and reset with each question (vendor-reported) |
| Access periods | 3, 6 and 12 months, plus a 48-hour free trial (vendor-reported) |
| Price | approx. £89.99 / £149.99 / £179.99 for 3 / 6 / 12 months (vendor-reported, 19 July 2026 — verify on the product page) |
Two things matter for this audit. First, the AKP bank is genuinely AKP-specific — you are not being sold a generic paediatric pool relabelled. Second, the AI layer is real but thinly documented: Pastest describes Tutor mode as on-demand help to clarify a point, break down answer options and explore a topic in more depth, and says it resets with each question. It does not, on the public pages, name the underlying model or describe how its answers are grounded. That single gap frames everything that follows, and it is the reason this is an audit rather than a recommendation.
The AKP exam anchor
Calibrate against the RCPCH format, not the marketing. Applied Knowledge in Practice is the second written stage of the MRCPCH, sat after Foundation of Practice and Theory and Science and before the MRCPCH Clinical (OSCE). It is 120 single-best-answer questions delivered as two papers of 60, two hours 30 minutes each, both on the same day, computer-based at an exam centre or online. There is no negative marking, so a blank is a wasted mark — you answer everything, even your best guess. That gives you roughly two and a half minutes an item, which sounds generous until you meet the paper's defining feature: AKP is applied. Expect clinical case scenarios, images, growth and observation charts, and data interpretation, all pitched at the level of a trainee moving through core paediatric training.
The syllabus spans the full paediatric curriculum — neonatology, cardiology, respiratory medicine and ENT, gastroenterology, nephro-urology, neurology and neurodevelopment, endocrinology and metabolism, haematology and oncology, infection and immunology, dermatology, musculoskeletal medicine, adolescent health, emergency medicine, behavioural medicine, genetics, nutrition and palliative care — wrapped around cross-cutting themes of safeguarding, patient safety, ethics and law, pharmacology and the science of practice. The RCPCH publishes the syllabus and a set of official example questions; those examples, not any vendor's, are your fidelity gold standard. Third-party banks approximate the blueprint; only the RCPCH defines it.
How we tested the Tutor mode
To audit an AI tutor you need a fixed, repeatable probe, not a general impression. We ran six representative AKP item types through the tutor and scored each answer against the same rubric. You can reproduce this on your own subscription in about twenty minutes.
| Probe | What it tests |
|---|---|
| Straight recall | A single-fact item, such as an inheritance pattern |
| Diagnosis | A vignette pointing to one best diagnosis |
| Next investigation | The correct next test, not the definitive one |
| Management | The best next step given a UK paediatric context |
| Ethics / safeguarding | A consent, capacity or safeguarding judgement |
| Ambiguous item | A deliberately under-specified stem |
Score each answer 0–2 on five axes: correctness against a primary source; grounding (does it point to a checkable authority?); reasoning (does it surface discriminating features, not just the answer?); restraint (does it flag uncertainty rather than bluff?); and fidelity (UK paediatric context, right level, right terminology). Publish your own scores rather than trusting ours; the point is to hold the tutor to a written standard, not to be charmed by fluent prose. For the fuller version of this method, we keep a standing guide to auditing an AI medical-exam tutor for grounding, answer leakage and hallucination.
Grounding audit: can you see where the answer comes from?
This is the crux. On the public product pages Pastest does not state which model powers Tutor mode or how its output is grounded, and the feature resets each question, so it carries no memory you can interrogate across a session. In practice that means an explanation may be excellent, but you cannot see whether it is drawn from Pastest's own peer-reviewed explanation, from a current NICE or CKS guideline, or generated on the fly. Treat any AI answer as an uncited claim until you have matched it to a named authority. For a paediatric prescribing or dosing point, that authority is the Summary of Product Characteristics (SmPC/eMC) for the specific medicine, plus the relevant NICE or CKS guidance — never the tutor's remembered number. Provenance you cannot inspect is provenance you must supply yourself.
You can audit provenance without copying anything proprietary: ask the tutor to name the guideline and its publication date, then check that guideline directly. If it cannot name one — or names one that does not say what it claims — that is your finding, and it should lower your trust in every unsourced sentence around it.
Reasoning behaviour: does it teach or does it tell?
The best exam tutors delay the answer and make you reason; the worst hand it over and let recognition masquerade as understanding. In our probes, Tutor mode was strongest as an on-demand explainer: it broke down why distractors were wrong and surfaced discriminating features when asked directly. It was weakest at Socratic restraint. Because it is designed to help you "while you answer", it will readily confirm or reveal, which is convenient and pedagogically risky — the moment it tells you the key, the item stops being a test. It handled a clean false premise reasonably when challenged ("are you sure the first-line investigation is X?"), but, like most general models, it can also fold too quickly and agree with a wrong correction. The lesson is procedural: the tutor's usefulness depends almost entirely on when you open it.
Exam fidelity: jurisdiction, terminology, timing and blueprint
AKP is a UK paediatric exam, and fidelity failures cluster in four places. Jurisdiction: an AI may quote non-UK immunisation schedules, drug brand names or thresholds; AKP expects the current UK schedule and UK guidance. Terminology: it should say adrenaline not epinephrine, paracetamol not acetaminophen, salbutamol not albuterol. Timing: the tutor is untimed, so it can train a leisurely, look-it-up habit that betrays you at two and a half minutes an item — worse, some of that budget is eaten by reading images and charts. Blueprint: a tutor answers the item in front of you; it does not tell you that you have done 400 cardiology items and barely touched neonatology or safeguarding. Coverage is a property of your whole attempt history, not of any single explanation, which is exactly why completion is not the same as coverage.
Five failure modes to watch
- Hallucinated citations — a named guideline that does not say what the tutor claims, or does not exist. Verify the source, not just the sentence.
- Overconfident wording — fluent, absolute phrasing on a genuinely contested point. Confidence is a style, not evidence.
- Outdated guidance — paediatric thresholds, resuscitation algorithms and immunisation schedules change; an AI may quote a superseded version as current.
- Answer leakage — the tutor reveals the key before you have committed, quietly converting a test into a read.
- Plausible-but-unexamined elaboration — extra "teaching" that sounds authoritative but was never on the blueprint and may simply be wrong.
A safe-use protocol: answer first, interrogate second, verify third
One routine neutralises most of those failure modes. Answer first: commit to your option and a one-line rationale before you open Tutor mode. This preserves the item as a real test and gives you a prediction to check the explanation against. Interrogate second: ask the tutor to argue the case, not state the verdict — "What are the two discriminating features between the best answer and the runner-up?" and "What would have to be different in the stem for the answer to change?" Verify third: for any management, dosing or threshold claim, close the tutor and confirm against a primary UK source — NICE, CKS, SIGN, the SmPC/eMC, or the RCPCH syllabus example. Only then log the item as learned. The order is the whole trick: reversing it turns a tutor into a spoiler.
Worked example: a seven-day plan around clinical work
Here is one honest week for a trainee on a busy rota, using Pastest for one defined job — drilling and explaining — and iatroX for a different one: unseen, timed measurement. No claim is made about either platform's internal algorithm.
- Monday (post-take, 30 min): 20 Pastest AKP items with Tutor mode OFF. Flag every miss.
- Tuesday (40 min): rework Monday's misses with Tutor mode ON, using the interrogate-second prompts; verify two dosing or threshold claims against the SmPC/eMC and CKS.
- Wednesday (25 min): read one Pastest textbook topic covering your weakest domain; write three of your own recall questions from it.
- Thursday (40 min): a fresh 30-item Pastest set concentrated in your two weakest blueprint areas.
- Friday (30 min): 20 unseen, timed iatroX items with no assistance, as a clean transfer check; record first-attempt accuracy only.
- Saturday (20 min): update your misconception log; re-test five items you got wrong last week.
- Sunday: rest, or a single timed 60-item mock if you are within a month of the exam.
The split matters. Pastest does the teaching; iatroX supplies the unseen, timed sample that tells you whether the teaching transferred — the essence of the two-Q-bank rule. Because iatroX's UK-core banks are free, it is a natural measurement bank that does not duplicate your Pastest drilling or inflate your percentage with re-seen items.
Decision checklist: continue, supplement, switch or stop
- Continue if your first-attempt accuracy on unseen, timed items is rising and your misses cluster in named, shrinking topics.
- Supplement with a second, non-duplicating measurement bank if your Pastest percentage is high but your unseen score lags — a classic sign you are recognising items, not mastering topics.
- Switch only for a measurable coverage gap — a domain the bank serves thinly that keeps appearing in your errors — never for novelty or a shinier interface.
- Stop a resource that is not changing your unseen score after two honest weeks; sunk cost is not a study plan.
The bottom line
Pastest is a strong, AKP-specific bank, and Tutor mode is a genuinely useful on-demand explainer — provided you treat it as an explainer, not an oracle. Its questions and human explanations are its real value; its AI adds convenience but hides its provenance, so the discipline of answer-first, interrogate-second, verify-third is non-negotiable. Buy it for the bank, use the AI on a leash, and measure your readiness on unseen items elsewhere.
Frequently asked questions
Is Pastest enough for MRCPCH Applied Knowledge in Practice on its own? For most candidates it is a sufficient primary bank: a 2,000-plus AKP-specific question pool (vendor-reported, 19 July 2026), human explanations and three past papers cover the blueprint broadly. "Enough", though, depends on transfer, not volume — you also need unseen, timed practice and the official RCPCH example questions to confirm the learning generalises to a fresh paper. As a single learning bank, yes; as your only source of readiness evidence, no.
Which MRCPCH Applied Knowledge in Practice component does Pastest not reproduce well? The thing no static bank reproduces is the live, mixed, image-and-data-heavy pressure of two 60-item papers at two and a half minutes an item in one sitting. Tutor mode is untimed and helps "while you answer", which trains a look-it-up rhythm the real AKP will not allow. Use Pastest to learn the content; simulate the timed, unassisted paper separately before exam day.
How should I verify Pastest AI answers for MRCPCH Applied Knowledge in Practice? Commit to your own answer first, then ask the tutor to name the guideline it relies on and its date, then check that guideline yourself — NICE, CKS, SIGN, the SmPC/eMC for any drug or dose, or the RCPCH syllabus example. If the tutor cannot name a checkable source, downgrade its answer to a hypothesis and confirm it against a primary reference before you commit it to memory.
When should I stop using Pastest and move to mixed mocks? When your first-attempt accuracy on fresh Pastest sets plateaus and your misses stop clustering, you have extracted most of the bank's teaching value; that is the trigger to shift weight toward full-length, mixed, timed mocks under exam conditions. In practice, over the final three to four weeks, prioritise whole 60-item timed papers over topic-by-topic drilling.
How should I combine Pastest with iatroX without duplicating practice? Give each a single, different job. Pastest is your primary learning bank — drilling, explanations and its AI tutor for rework; iatroX is your measurement bank — small sets of unseen, timed, mixed items with no assistance, read for first-attempt accuracy only. Keeping the two roles separate is exactly the two-Q-bank rule, and it stops you re-testing questions you have already seen and mistaking a rising percentage for readiness.
Editorial notes and references
Written by Dr Kolawole Tytler, NHS GP and founder of iatroX. Last checked 19 July 2026; vendor-reported figures (question counts, prices, access periods and the Tutor mode description) were accurate to the Pastest AKP product page on that date and change without notice — verify the current figure on the product page before relying on it. Disclosure: iatroX operates a competing UK question bank and clinical-knowledge platform; this audit confines iatroX's role to jobs Pastest does not itself claim — unseen readiness measurement and Socratic rework of missed items — and does not position iatroX as a replacement for a primary content bank. Corrections are welcome via the feedback route on iatrox.com.
References: RCPCH — Theory exams: structure and syllabi (rcpch.ac.uk/education-careers/examinations/theory/structure-syllabi) and the official AKP example questions; Pastest MRCPCH AKP (pastest.com/products/mrcpch-akp); iatroX — "Your Q-Bank Percentage Is Not Your Exam Score"; "How to audit an AI medical-exam tutor"; the two-Q-bank rule; and the iatroX comparison hub.
Test a fresh, unseen MRCPCH AKP question in iatroX and verify before you reveal →