Pastest is an established MRCEM provider, and its Tutor mode adds on-demand AI to a large emergency-medicine bank — useful in exact proportion to a property you cannot see on the sales page: whether its explanations are grounded in checkable sources or generated on the fly. This audit gives MRCEM SBA candidates a reproducible way to find out and a protocol for using the tutor without inheriting its errors. The principal limitation: emergency medicine turns on current UK resuscitation and acute-care guidance, and confident AI output is least reliable exactly where currency and jurisdiction matter most.
What Pastest offers for MRCEM SBA right now
Pastest's MRCEM Intermediate SBA preparation follows its platform model: an exam-specific single-best-answer bank, past-paper-style practice, progress analytics, and Tutor mode's instant AI support during questioning, behind tiered fixed-term pricing. Verify current MRCEM SBA counts and price on the product page. Pastest's editorial base and long presence in emergency-medicine revision are real strengths; this audit concerns the generated layer over it and how it behaves against an RCEM-curriculum exam.
The exam that sets the bar
The MRCEM SBA is 180 single-best-answer questions in two 120-minute papers of 90 questions each, mapped to the 2021 RCEM curriculum across Specialty Learning Outcomes 1 and 3–7, weighted Complex Stable Patient 55, Resuscitation 40, Injured Patient 30, Paediatric Emergency Medicine 25, Procedural Skills 20 and Complex/Challenging Situations 10. Three properties stress an AI tutor: the content standard is UK emergency practice (Resuscitation Council UK algorithms, NICE, RCEM guidance); resuscitation and acute-care answers move with guideline updates; and the paper is fast (80 seconds an item), so the skill is quick discrimination, which an always-available explainer can erode.
The audit rubric: six item types, four dimensions
Run six item types through Tutor mode and score each 0–2 on four dimensions, for a rerunnable 48-point audit. Items: a recall item (a dose, a threshold); a diagnosis vignette; a next-investigation item; a management item where UK resuscitation or NICE guidance is specific; an ethics or safety item; and one ambiguous item where the "single best" answer is arguable. Dimensions: grounding (traceable to Pastest's own explanation or a named UK guideline, or free-floating?), reasoning (engages your logic or restates the key?), calibration (confidence drops on ambiguous items?), fidelity (UK emergency terminology, UK algorithms, RCEM-curriculum awareness).
Grounding: make it show its UK sources
For any explanation that will change how you answer future questions, ask: "What UK guideline supports this, and what is its date?" A tutor standing on Pastest's expert-written explanations is on checkable editorial ground; one naming a Resuscitation Council UK algorithm, NICE guidance or an RCEM standard with a date is verifiable in a minute; one producing confident, sourceless prose is unverified — and in emergency medicine, where algorithms and thresholds update, unverified fluency is where stale advice hides. Log the ratio across your six-item audit and re-run it quarterly.
Reasoning behaviour: the tests that take ten minutes
Demand engagement with your stated reasoning, near-miss discrimination (why the best distractor fails), and honest uncertainty on contested items. Then run the false-premise test: assert something plausible and wrong — an outdated dose or a superseded threshold — and see whether the tutor corrects you cleanly or agrees. A tutor that fails that test will fail it silently, always pleasantly, which is dangerous in a specialty where a confidently-wrong dose is a patient-safety issue as well as an exam error.
Exam fidelity
Probe jurisdiction with a resuscitation or acute-care question where UK and international practice may differ, and check both the answer and whether it reflects current UK algorithms. Probe curriculum-awareness: does the tutor connect explanations to the SLO domains the SBA samples, or wander into subspecialty detail the exam will not reward? Probe pace-realism: explanations that train elaborate deliberation are luxury goods when the paper allows 80 seconds an item.
The failure modes that matter
Log five deliberately: hallucinated or unverifiable citations; overconfident wording on contested points; outdated guidance delivered fluently (the highest-probability failure in a guideline-driven specialty); answer leakage that erodes commit-first practice; and plausible elaboration. Weekly, verify five random outputs against primary UK sources and log discrepancies. In emergency medicine, add a sixth watch item: confidently-stated doses and thresholds, which deserve verification every time because the harm of an error is highest.
The safe-use protocol
Answer first — option and one-line rationale committed before the tutor opens. Interrogate second — why was my reasoning wrong, what single feature discriminates my answer from the key, what stem change flips it? Verify third — anything behaviour-changing, and every dose or threshold, gets checked against a named UK source, or put through a system built to answer with UK citations: Ask iatroX grounds its answers in NICE, CKS, SIGN, SmPC/eMC and NHS content. Weekly, run the five-output verification sample. Trust the log, not the fluency.
A seven-day pattern for busy trainees
Monday: 40 Pastest questions in the complex-stable-patient or paediatric-EM domains, tutor closed until after commitment. Tuesday: 30 questions plus review of resuscitation and acute-care currency. Wednesday: a timed, unseen 50-question mixed block in iatroX's free MRCEM SBA bank, no tutor, at 80 seconds an item. Thursday: misconception-log review; weekly verification sample. Friday: 40 questions on flagged SLO domains, timed. Saturday: full timed simulation, same-day review by error type. Sunday: rest. Pastest explains on demand; iatroX measures on unseen material and verifies with citations; your reasoning stays the integration layer.
Continue, supplement, switch or stop
Continue if the rubric scores well and your verification log stays clean — especially on doses and currency. Supplement when explanation quality is high but unseen timed performance is flat. Switch only on logged failure patterns — repeated stale algorithms, confident dose errors. Stop all AI assistance in final-fortnight simulations: the SBA arrives without an assistant.
Reading your audit scores
Read the 48-point rubric by dimension, not as a total. High grounding with low reasoning is an accurate but pedagogically flat tutor — verify with it, reason for yourself. Low grounding with high reasoning is the profile to fear in emergency medicine, where a fluent voice without checkable sources delivers confidently-wrong doses and superseded algorithms; force a named, dated UK source on every behaviour-changing claim from a tutor scoring this way, and verify every dose and threshold regardless. Calibration that stays honest on ambiguous items is the marker of a trustworthy tool; a fidelity score that sags on jurisdiction means it is good general medicine aimed at the wrong resuscitation standard.
Why doses and algorithms deserve zero-tolerance verification
Emergency medicine concentrates two features that make AI errors especially costly: time-critical decisions and precise numeric answers. A confidently-stated but outdated adrenaline dose, a superseded thrombolysis threshold, or a wrong paediatric weight-based calculation is both an exam error and, in the clinician the exam is certifying, a patient-safety issue. Large language models are known to state such specifics fluently and sometimes wrongly, and the fluency is exactly what makes the error hard to catch. So for this exam, add a rule beyond the standard protocol: every dose, threshold and algorithm the tutor states gets checked against a current UK source before you internalise it, every time, no exceptions. The harm of a wrong number here is high enough that the few seconds of verification are never wasted, and the habit you build — never trust a stated dose without checking — is one worth carrying past the exam into practice.
Frequently asked questions
Is Pastest enough for MRCEM SBA on its own? Its bank and AI can anchor preparation, but you still need an SLO-weighted coverage audit, unseen timed blocks, and the verification habit this audit describes — and an always-available explainer can undermine the retrieval practice SBA performance is built on unless you commit first.
Which MRCEM SBA component does Pastest not reproduce well? The exam's SLO-weighted breadth under unseen timed conditions, and the fast pace; an AI tutor also cannot reproduce retrieval practice and, used carelessly, erodes it.
How should I verify Pastest AI answers for MRCEM SBA? Require a named, dated UK source for any behaviour-changing claim, verify every dose and threshold, check via a citation-grounded system such as Ask iatroX, and audit five random outputs weekly with a written log.
When should I stop using Pastest and move to mixed mocks? Once SLO coverage is complete, accuracy is stable for a fortnight and pace fits the 80-second budget — the final two to three weeks, tutor closed during blocks.
How should I combine Pastest with iatroX without duplicating practice? Pastest for commit-first drilling and reasoning prompts; iatroX (free for MRCEM SBA) for unseen adaptive measurement across the SLO domains, Socratic repair and cited UK-guideline verification.
The bottom line for busy trainees
The honest one-line verdict on Pastest's AI Tutor for the MRCEM SBA: useful in proportion to how strictly you verify, because emergency medicine concentrates the two things AI gets confidently wrong — time-critical decisions and precise numbers. A fluent but outdated dose, a superseded algorithm or a wrong weight-based calculation is both an exam error and, in the clinician the exam certifies, a patient-safety issue, and the fluency is exactly what makes it hard to catch. So beyond the standard commit-interrogate-verify routine, add a zero-tolerance rule: every dose, threshold and algorithm the tutor states gets checked against a current UK source before you internalise it, every time. Do that and Pastest's AI is a genuine asset on a strong emergency-medicine bank; skip it and the tutor becomes a confident source of exactly the stale specifics the SBA — and your future patients — will penalise.
A final calibration note: re-run the six-item audit each time Pastest updates its AI, which can happen silently between sittings, because a tutor that grounded well last term can regress after an unannounced model change. In a guideline-driven specialty like emergency medicine, that quarterly re-check is not optional diligence — it is the difference between trusting a tool you have verified and trusting one you verified months and several algorithm updates ago, on an exam where a stale answer can be a patient-safety issue as well as a lost mark.
Editorial notes and references
Written by Dr Kolawole Tytler, NHS GP and founder of iatroX. Last checked 19 July 2026; Pastest features are vendor-published and evolving — confirm MRCEM SBA counts, price and AI capabilities on the product page. MRCEM SBA format is per RCEM. Disclosure: iatroX operates a free competing MRCEM SBA bank and a Socratic Tutor; run this rubric against ours with equal severity. Corrections via the feedback route on iatrox.com. References: RCEM MRCEM exam and curriculum pages (rcem.ac.uk); Pastest MRCEM pages (pastest.com); related reading: why your Q-bank percentage is not your exam score.