This audit is for MRCEM Intermediate SBA candidates deciding how far to trust FRCEMtutor's analytics — and it has to correct the question in the title before answering it. On the evidence of its public pages on 19 July 2026, FRCEMtutor does not appear to run an algorithmic "adaptive engine" in the sense of a feed that auto-selects your next question. What it offers is genuine and useful — a large Intermediate-mapped bank, performance analytics, peer benchmarking and user-configured tests — but the personalisation is largely something you drive, not an algorithm optimising on your behalf. That distinction changes how you should read every number it shows.
What FRCEMtutor offers for MRCEM SBA right now
Current-state box — last checked 19 July 2026; figures vendor-reported and subject to change:
| Item | What we found (vendor-reported, 19 Jul 2026) |
|---|---|
| Intermediate SBA bank | "over 1250 SBA questions" mapped to the 2021 RCEM syllabus (one page states "over 1000") |
| Also offered | MRCEM Primary bank ("over 3500 SBA questions" plus a knowledge tutor) |
| Access / price (Intermediate) | 2 months £20 · 3 months £25 · 4 months £30 · 6 months £35 |
| "Adaptive"/AI engine | None described; no algorithmic auto-selection or AI stated on the pages reviewed |
| Personalisation | "Individual learning tailored to your needs" via user-configured timed tests (choose number of questions and time limit) |
| Analytics | Performance analysis, peer benchmarking against fellow candidates, and a candidate performance graph |
| Other | Reference text with user annotations; timed-test mode |
Unlike some competitors, FRCEMtutor genuinely covers the Intermediate SBA, and its Intermediate pricing is modest. Verify each figure on the FRCEMtutor Intermediate SBA page; counts and prices change.
The honest finding on "adaptive"
The article's premise deserves a straight answer: what is FRCEMtutor's adaptive engine optimising? On the pages we reviewed, we found no description of an adaptive learning engine, artificial intelligence, or algorithmic personalisation. The "personalised development" on offer is you choosing question counts, time limits and topics, plus analytics and peer comparison to guide those choices. That is not a criticism — user-configured practice with good analytics is a perfectly sound way to revise — but it means there is no hidden algorithm quietly steering your coverage, for better or worse. The optimisation is yours to do. If you want the low-volume domains protected, you must protect them; nothing does it automatically. Always confirm on the live product page in case features have changed since this check.
Exam anchor: the Intermediate SBA blueprint
The MRCEM Intermediate SBA is 180 single-best-answer questions in two 120-minute papers of 90 each, against the 2021 RCEM curriculum (SLOs 1 and 3–7), no negative marking, roughly 80 seconds per item, modified-Angoff pass mark typically 60–70%. The domain weighting:
| RCEM domain (SLO) | Approx. items | Share |
|---|---|---|
| Complex stable patient (SLO 1) | 55 | 31% |
| Resuscitation / acutely unwell (SLO 3) | 40 | 22% |
| Injured patient (SLO 4) | 30 | 17% |
| Paediatric EM (SLO 5) | 25 | 14% |
| Procedural skills (SLO 6) | 20 | 11% |
| Complex/challenging situations (SLO 7) | 10 | 6% |
A vendor's own topic categories are a claim; audit them against this official weighting.
Define every metric before you trust it
FRCEMtutor's dashboard is only as useful as your grip on what each figure means:
- First-attempt accuracy — the only accuracy that reflects genuine recall; insist on it over blended figures.
- Repeat accuracy — measures memory of seen items, not understanding.
- Percentile / peer benchmark — your rank against other FRCEMtutor users sitting the exam, a self-selected, unverified cohort; motivating, but not a pass probability.
- Predicted score — if shown, a modelled estimate only as good as its calibration; never treat it as a guarantee.
- Coverage — proportion of the bank attempted (completion), not blueprint coverage; a 100%-complete bank can still under-represent an SLO.
- Difficulty — usually historical percentage-correct across users, which drifts with the user base.
- Time per item — useful for pacing against the ~80-second budget, but only in timed, unassisted mode.
Selection bias: why the percentage is not the score
Even without an algorithm oversampling your weak spots, self-directed practice creates its own bias: candidates drift towards comfortable topics and re-do items they half-remember. A rising overall percentage built on filtered or repeated questions is not comparable to a score on a mixed, unseen, blueprint-proportioned block. This is why the only number worth acting on is earned on fresh items — a point made in full in your Q-bank percentage is not your exam score. Peer benchmarking adds a second trap: being at the 70th percentile of a self-selected cohort tells you about that cohort, not about the passing standard.
Blueprint audit beats the home-screen average
Rebuild your attempted distribution against the RCEM weighting instead of trusting the average. A worked example three weeks out:
| Domain | Official share | Your attempts | First-attempt acc. | Verdict |
|---|---|---|---|---|
| Complex stable patient | 31% | 41% | 70% | Over-weighted; strong |
| Resuscitation | 22% | 24% | 61% | On-blueprint |
| Injured patient | 17% | 14% | 57% | Slightly under; weak |
| Paediatric EM | 14% | 7% | 48% | Under-sampled; weak |
| Procedural skills | 11% | 8% | 55% | Under-sampled |
| Complex/challenging | 6% | 6% | 45% | On-volume; weak accuracy |
The home screen might show a reassuring 63% and a flattering peer percentile. The audit shows paediatrics and procedures under-sampled and the complex/challenging domain weak — none of which the average reveals, and none of which an absent algorithm will fix for you.
The readiness test: five conditions
A percentage is a readiness signal only when it is unseen, timed, mixed, unassisted and adequately sampled (a full 90-question paper, ideally two). Fail any one and the figure is feedback, not readiness. Because FRCEMtutor's analytics are computed on items you will increasingly have seen, you need a separate unseen stream to satisfy the "unseen" condition honestly.
Algorithm override rules — which here means you
Since nothing auto-selects for you, set explicit manual overrides so the small, high-consequence areas are practised:
- Paediatric EM — the unwell child, weight-based dosing, safeguarding.
- Procedural skills — indications, contraindications, complications.
- Complex/challenging situations — capacity, consent, end-of-life, major incident.
- Images and data — ECGs, radiographs, gases and laboratory trends.
- Calculations — infusions, corrected values, paediatric weights.
Build these into your weekly timed tests deliberately; the analytics will then tell you whether the override is working.
Worked dashboard example: from analytics to quotas
Convert the audit into next week's quotas without a spurious prediction. With ~450 minutes of study:
- Paediatric EM — 60 unseen items, timed (largest gap).
- Procedural skills — 40 items, timed.
- Complex/challenging — 25 items plus a short read on mental-capacity assessment.
- Injured patient — 30 items to close the small gap.
- One mixed, timed 90-item block at week's end as the only readiness figure logged.
A seven-day pattern for busy trainees
| Day | FRCEMtutor job | iatroX job |
|---|---|---|
| Mon–Wed | Configured timed tests on your two weakest SLOs, 40–50 items/day; review by cause | — |
| Thu | Rework errors; annotate the reference text; no new items | — |
| Fri | — | Fresh, timed, unseen mixed block in iatroX; score once and log |
| Weekend | Short read on Friday's weakest domain; rest | — |
FRCEMtutor is the learning-and-analytics engine; iatroX is the unseen measurement layer. No claim is made to any proprietary algorithm on either side.
Three mistakes this audit is designed to stop
Mistake one: assuming an algorithm has your back. FRCEMtutor's personalisation is user-driven; if you do not schedule the small SLOs, nothing else will.
Mistake two: reading the peer percentile as a pass probability. It ranks you against a self-selected cohort, not against the standard.
Mistake three: trusting analytics computed on seen items. Repeat accuracy and rising averages measure memory; only unseen blocks measure readiness.
Decision checklist: continue, supplement, switch or stop
- Continue with FRCEMtutor if its Intermediate bank, analytics and peer benchmarking are actively closing labelled gaps and you are driving the coverage yourself.
- Supplement with an unseen measurement bank as soon as you need an honest readiness figure its own analytics cannot give.
- Switch primary bank only for a measurable coverage gap, not because you expected an adaptive engine and found analytics instead — reset the expectation, keep the bank if it covers the blueprint.
- Stop new questions when unseen, timed, blueprint-proportioned blocks are stable and official-material calibration is done.
Bottom line
FRCEMtutor is a legitimately useful, Intermediate-mapped, affordably priced bank with sound analytics and peer benchmarking — but on the evidence reviewed it is not an adaptive algorithm, so the coverage discipline is on you. Audit your distribution against the RCEM weighting, protect the low-volume SLOs by hand, and measure readiness on unseen blocks. Compare options on the iatroX comparison hub rather than on peer percentiles.
FAQ
Is FRCEMtutor enough for MRCEM SBA on its own? FRCEMtutor covers the Intermediate SBA with a sizeable, syllabus-mapped bank (vendor-reported "over 1250" questions, 19 July 2026) and useful analytics, so it can serve as a candidate's main engine. It is not "enough" in the sense that its analytics run on items you will have seen, so you still need a separate unseen stream for an honest readiness figure, and its personalisation will not protect the small domains unless you configure it to.
Which MRCEM SBA component does FRCEMtutor not reproduce well? As a written-SBA bank it does not rehearse the subsequent MRCEM OSCE, and within the SBA its coverage of the low-volume, high-consequence domains — paediatric EM, procedural indications and the complex/challenging situations strand — depends on you selecting them, because there is no algorithm oversampling weak areas on your behalf. Image and calculation items also need deliberate inclusion.
How many FRCEMtutor questions should I complete per day for MRCEM SBA? A sustainable 40 to 50 well-reviewed items a day around shifts, front-loaded onto your weakest SLOs and error-labelled, is more valuable than a larger unfocused total. Because the platform will not auto-balance your coverage, spend some of that daily allocation deliberately on the small domains rather than the comfortable large one.
When should I stop using FRCEMtutor and move to mixed mocks? Shift towards timed, mixed mocks once your weekly domain floors are cleared, first-attempt accuracy across SLOs is converging and pacing is reliable at ~80 seconds per item. Keep some configured practice for residual weak spots, but let mixed unseen blocks dominate the run-in, since bank completion and a rising peer percentile are not readiness signals.
How should I combine FRCEMtutor with iatroX without duplicating practice? Assign non-overlapping jobs: learn and drill by topic on FRCEMtutor, and measure on unseen, blueprint-mapped items in iatroX that you have not seen before. Keeping learning and measurement on different banks means your readiness figure reflects recall rather than memory of FRCEMtutor's own questions — the practical form of the two-Q-bank rule.
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 feature descriptions, including the absence of an adaptive engine) were taken from FRCEMtutor's public pages on 19 July 2026 and change without notice — verify the current figures and features on the product page before purchasing. Disclosure: iatroX operates a competing MRCEM SBA question bank; this audit confines iatroX to unseen, blueprint-mapped measurement — a job FRCEMtutor's analytics cannot do on seen items — and speaks positively of FRCEMtutor where its coverage and pricing warrant it. Corrections: use the feedback route on iatrox.com and we will amend any error or out-of-date figure.
References