Pastest now ships an AI assistant — Tutor mode, "instant, on-demand AI support while you answer questions" — inside one of the largest MRCP Part 1 banks on the market. This audit is for candidates deciding how much to trust it. The method is a reproducible rubric you can run on your own subscription, and the principal limitation to hold onto is that an AI explanation layer is only as safe as its grounding, and grounding is exactly what vendors document least.
What Pastest offers for MRCP Part 1 right now
Checked against the vendor's public pages on 19 July 2026: over 5,000 exam-style questions and more than 30 bespoke past papers, with analytics that promise "real-time insights into your progress and performance", nearly 700 textbook topics, and Tutor mode's AI support available while answering. Access is sold at £99.99 for three months, £144.99 for six and £179.99 for twelve, with a 48-hour free trial. Pastest's question volume, past-paper framing and multimedia depth are real strengths and the reason it has survived every wave of competition; none of what follows disputes that. The question this audit asks is narrower: when the AI layer explains, what is the explanation standing on?
The exam that sets the bar
MRCP(UK) Part 1 is two three-hour papers of 100 best-of-five questions, no negative marking, with a published blueprint that concentrates almost a fifth of the exam in clinical sciences and clinical pharmacology and spreads the rest across the organ specialties (14 questions each for the major ones, down to 4 for ophthalmology and palliative medicine). Two properties matter for an AI tutor. The distractors are engineered to be nearly right, so an explanation that cannot articulate why the near-miss option fails is not doing the exam's work. And the content standard is UK practice — an assistant that drifts into another jurisdiction's first-line choices is training you for someone else's exam.
The audit rubric: six question types, four dimensions
Run a fixed set of six item types through Tutor mode and score every response on the same four dimensions. The item types: a straight recall item (a receptor, an enzyme, a mechanism); a diagnosis vignette; a next-investigation item; a management item where UK guidance is specific; an ethics or consent item; and one genuinely ambiguous item where reasonable clinicians could argue. The dimensions: grounding (does it cite or clearly derive from an identifiable source?), reasoning (does it engage with your thinking or just restate the answer key?), calibration (does it express uncertainty where uncertainty exists?), and fidelity (is the answer true to MRCP Part 1's UK context and depth?). Score each 0–2 and you have a 48-point audit you can rerun each quarter — and compare honestly with any other tutor, including ours.
Grounding: where do the answers come from?
The provenance question has three possible answers, in descending order of safety: the tutor cites Pastest's own expert-written explanation and enriches it; it cites an external guideline by name and date; or it produces fluent, uncited model output. You can distinguish these without reproducing any copyrighted content: ask "what is the source and date of the guidance behind this answer?" and watch whether the reply names something checkable or gestures at "current guidelines". A named, dated source lets you verify in sixty seconds. An unnamed one converts every explanation into a claim you are accepting on the vendor's behalf. Log what you get across your six test items; the pattern, not any single response, is the finding.
Reasoning behaviour: assistant or answer key?
The pedagogically valuable behaviours are specific: asking what you thought before telling you; engaging with your differential rather than the model answer only; saying "this is contested" when it is; and pushing back cleanly when you assert something false. The failure behaviours are equally specific: revealing the answer the moment you open it (which converts retrieval practice into recognition practice); agreeing with a false premise you smuggle into your question — test this deliberately with a plausible wrong claim like "since beta-blockers are first-line monotherapy for essential hypertension in the under-55s…"; and hedging so uniformly that nothing is ever actually asserted. Ten minutes of adversarial prompting tells you more about a tutor than any feature list.
Exam fidelity: is it revising you for this exam?
Probe three axes. Jurisdiction: ask a management question where UK and US practice diverge and see which answer arrives, and whether the divergence is flagged. Depth: Part 1 rewards mechanism-level understanding — an explanation pitched at final-year recall is not wrong, just not this exam. Weighting: does the assistant know that clinical sciences and pharmacology dominate the blueprint, or does it treat every topic as equally examinable? An AI layer that is excellent general medicine but indifferent to the specific exam will feel helpful daily and cost you marks quietly.
The failure modes that matter
Five recur across every AI explanation layer in this class, and Pastest's should be tested against each: hallucinated or unverifiable citations (a guideline named confidently that does not say what is claimed); overconfident wording where the evidence is genuinely uncertain; outdated guidance delivered fluently — the failure mode that ages worst, since UK guidance moves annually; answer leakage that quietly deletes the testing effect your subscription exists to generate; and plausible elaboration — extra "facts" appended to a correct answer that nobody wrote and nobody checked. None of these is unique to Pastest, and their presence in some form is close to certain in any current LLM product. What separates platforms is frequency and detectability, which is what your logged rubric measures.
The safe-use protocol
Three steps, in order, every time. Answer first: commit to an option and a one-line rationale before opening the tutor — this is non-negotiable, because it preserves retrieval practice and gives the AI something real to correct. Interrogate second: ask why your reasoning failed, why the best distractor is wrong, and what single discriminating feature separates them — high-value prompts that produce transferable rules rather than restated answer keys. Verify third: for anything that will change how you answer future questions — and certainly anything that would change how you treat a patient — check the named source, or run the claim past a retrieval-grounded system that cites UK guidance directly, which is precisely the job Ask iatroX was built for. Once a week, take five tutor outputs at random and verify them against primary sources; log the discrepancies. If your log stays clean, your trust is earned; if not, you found out for the price of an hour.
A seven-day pattern that uses the AI without depending on it
Monday and Tuesday: 40 Pastest questions each day in your two weakest blueprint domains, tutor closed until after each commitment, interrogation reserved for genuine misses. Wednesday: a timed, unseen 50-question mixed block in iatroX's free MRCP Part 1 bank for a first-attempt readiness signal outside your Pastest history. Thursday: light review of the week's misconception log. Friday: 40 Pastest questions from a past paper, timed. Saturday: your weekly verification sample, then a second mixed block — Pastest past paper or iatroX adaptive session, alternating weeks. Sunday: rest. Pastest supplies volume, past-paper texture and on-demand explanation; iatroX supplies unseen transfer testing and source-grounded verification; the seams between them are where your own reasoning lives, which is the point.
Reading your audit scores
A completed 48-point rubric is easiest to act on if you read it by dimension rather than as a total. A high grounding score with a low reasoning score means the tutor is accurate but pedagogically lazy — it tells you the right answer with a checkable source but never engages your logic, so you should use it for verification and lean on your own error analysis for the reasoning work. The reverse — strong reasoning, weak grounding — is more dangerous, because a tutor that argues persuasively without checkable sources is exactly the profile that delivers fluent, stale guidance with conviction; treat its every management claim as unverified until sourced. A uniformly high calibration score is the quiet reassurance most people overlook: a tutor that reliably flags its own uncertainty is one you can trust to tell you when to go and check, which is worth more than raw accuracy. And a fidelity score that sags on the jurisdiction item is the single most important finding for a UK candidate, because it means the tool is competent medicine aimed at the wrong exam.
Continue, supplement, switch or stop
Continue if your rubric scores are solid and your verification log is clean — a grounded AI layer on top of Pastest's volume is a genuinely strong combination. Supplement if the tutor explains well but your unseen mixed performance is flat: explanation quality is not the bottleneck, retrieval practice is. Switch only on measurable failure — repeated unverifiable citations, jurisdiction drift you have logged — not on a competitor's launch video. Stop using any tutor, ours included, in your final fortnight's mock blocks: the exam has no assistant, and your last rehearsals should not either.
Frequently asked questions
Is Pastest enough for MRCP Part 1 on its own? Its question volume, past papers and textbook depth can anchor a full preparation, but its AI explanations need the verification habit this audit describes, and every candidate needs a source of unseen, timed, mixed practice outside their main bank to measure readiness rather than familiarity.
Which MRCP Part 1 component does Pastest not reproduce well? No commercial bank reproduces the exam's blueprint indifference — the forced coverage of clinical sciences, pharmacology and micro-domains under time pressure — and an on-demand explainer, used carelessly, actively erodes the retrieval practice on which Part 1 performance is built.
How should I verify Pastest AI answers for MRCP Part 1? Ask the tutor to name the source and date behind any answer that will change your future behaviour, check named sources directly, run a weekly random sample of five outputs against primary guidance, and route UK-guideline questions through a citation-first system such as Ask iatroX for a second, checkable opinion.
When should I stop using Pastest and move to mixed mocks? When your blueprint coverage is complete, first-attempt accuracy has been stable for two weeks and pacing sits within exam speed, spend the final three to four weeks predominantly on timed past papers and unseen mixed blocks, with the AI tutor closed during blocks and used only in review.
How should I combine Pastest with iatroX without duplicating practice? Split the jobs cleanly: Pastest for volume drilling, past papers and in-the-moment explanation; iatroX for unseen adaptive blocks that measure transfer, the Socratic Tutor when you want reasoning interrogated rather than answers supplied, and cited UK-guideline verification — two tools, two functions, no repeated questions.
Editorial notes and references
Written by Dr Kolawole Tytler, NHS GP and founder of iatroX. Last checked 19 July 2026; Pastest counts, prices and feature descriptions are vendor-published and were checked on that date — confirm on the product page before purchase, and expect AI features to evolve quickly. Disclosure: iatroX operates a competing MRCP Part 1 bank and a Socratic Tutor; the rubric above is deliberately platform-neutral so you can run it on our tutor too. Corrections via the feedback route on iatrox.com. References: MRCP(UK) Part 1 format and blueprint, thefederation.uk/examinations/part-1/format; Pastest MRCP Part 1 product pages, pastest.com; related reading: turning Pastest misses into a tutor session, the best MRCP Part 1 banks in 2026, and why your Q-bank percentage is not your exam score.