This audit is for international medical graduates using MedLumen's AMC CAT MCQ bank who want to know whether its analytics measure readiness or merely reward familiarity. MedLumen addresses the written AMC MCQ only — the single-best-answer paper, not the AMC Clinical Examination. The principal limitation is not the question count. It is that a self-selected, weak-area-weighted feed produces a headline percentage that is easy to over-read and hard to compare with a genuine mixed, unseen, timed block.
What MedLumen offers for the AMC MCQ right now
The figures below are vendor-reported from MedLumen's AMC CAT MCQ product page and were last checked on 19 July 2026. Treat them as a snapshot, not a contract: banks change counts, pricing and features without notice, so confirm every number on the live product page before you buy.
| Attribute | MedLumen for AMC CAT MCQ (vendor-reported, 19 July 2026) |
|---|---|
| Question volume | "732+" AMC practice questions; verify the current count on the product page |
| Access periods | 7-day free trial (no card required); three-month and twelve-month paid tiers |
| Price | Three-month and twelve-month tiers listed at 55 and 115 in the currency shown at checkout; verify the current price and currency on the product page |
| AI / adaptive study features | "AI-adaptive practice", "AI weak-area tracking", an "AI analytics dashboard" and "discipline analytics" |
| Assessment mode | "Full-length", "timed, mixed" mock exams |
| Localisation | States that each answer "explains the management expected in Australian practice" |
| Components supported | The AMC CAT MCQ written paper only — not the AMC Clinical Examination |
Two things deserve an early flag. First, MedLumen describes the AMC exam itself as "computer-adaptive" with "difficulty [that] adapts to your performance." The AMC's own published specification does not confirm an adaptive-difficulty algorithm; the exam is named a Computer Adaptive Test but the specification describes a fixed, 150-item computer-administered paper. Treat any vendor claim that the real exam changes difficulty as unverified, and prepare as if every candidate sees a fixed-length paper. Second, MedLumen is a multi-exam international bank (PLAB, USMLE, MRCP and AMC), so its Australian localisation depth is the thing to test hardest, not its interface.
The exam you are actually preparing for
The AMC MCQ is 150 single-best-answer questions, each with one correct option of five, delivered in one 3.5-hour session, computer-administered at Pearson VUE test centres. Every item is scored. The AMC MCQ Specifications (V8, dated 9 September 2025) set out the content the paper samples: the disease process, clinical examination and diagnosis, investigation, and therapy and management, across a defined discipline blueprint. Passing it is the gateway to the AMC Clinical Examination on the standard pathway.
The discipline weighting is what any coverage analytics must be judged against:
| AMC MCQ discipline (per AMC MCQ Specifications) | Approximate share of the 150-item paper |
|---|---|
| Adult Health — Medicine | ~30% |
| Adult Health — Surgery | ~20% |
| Women's Health (obstetrics and gynaecology) | ~12.5% |
| Child Health | ~12.5% |
| Mental Health | ~12.5% |
| Population Health and Ethics | ~12.5% |
These proportions are drawn from the AMC MCQ Specifications; confirm the exact current percentages in the latest version, as the AMC revises the document periodically. The specification also notes that at least half the paper is drawn from previously calibrated items with the remainder new, and that the exam tests reasoning rather than recall. The distinction that matters for a MedLumen audit: the AMC defines its own weighting, and any third-party bank's home-screen average is a summary of your attempts, not of the blueprint.
What each MedLumen metric means — and what it does not
Analytics dashboards use a small, recurring vocabulary. Define each term before you trust it.
First-attempt accuracy is the percentage of questions you answered correctly the very first time you saw them. This is the only accuracy figure with diagnostic value, because it is unrehearsed. Repeat accuracy is your percentage on questions you have already attempted; it climbs naturally as you memorise items and tells you little about exam performance. Coverage is the proportion of the bank you have touched — completion, not competence; finishing the bank does not mean you have covered the blueprint. Difficulty is usually a per-item tag derived from how the whole user base performs, useful for spotting which topics genuinely stretch candidates. Time per item is your average pace; against a 3.5-hour, 150-question paper you have roughly 84 seconds per item, so anything drifting past 90 seconds on straightforward recall is a warning. Percentile ranks you against other users, and predicted score projects a future result — MedLumen's page confirms discipline analytics and weak-area tracking, but I could not confirm a published percentile or a validated predicted-score model, so if either appears in your dashboard, treat it with particular caution and verify how it is calculated on the product page.
The one figure to elevate above all others is first-attempt accuracy on unseen items. Everything else is context.
Selection bias: why an adaptive feed makes your percentage hard to read
An "AI-adaptive" study feed is designed to surface your weak areas more often. That is pedagogically sensible, but it quietly corrupts your headline percentage. If the engine keeps feeding you cardiology because you keep failing it, your recent accuracy is dragged down by a deliberately hard, non-representative sample — you look worse than you are. Conversely, once you have drilled a weak area to memorisation, repeat accuracy inflates the same number and you look better than you are. Neither figure is comparable with the real exam, which samples the whole blueprint at a fixed difficulty in a fixed order.
This is the core reason a weak-area-weighted percentage cannot be your readiness signal. To read your true standing you need the opposite of an adaptive feed: a blueprint-proportioned, unseen, mixed block, scored once. Your Q-bank percentage is not your exam score — it is a training artefact, and an adaptive one doubly so.
Blueprint audit: attempted distribution versus the AMC weighting
Do not trust the home-screen average. Export or read off your attempted-question counts by discipline and lay them beside the AMC weighting. The question is whether your effort is proportioned like the paper. A worked comparison:
| Discipline | AMC target share | Your attempted share (example) | Gap |
|---|---|---|---|
| Adult Health — Medicine | ~30% | 41% | Over-sampled |
| Adult Health — Surgery | ~20% | 17% | Slightly under |
| Women's Health (O&G) | ~12.5% | 8% | Under-sampled |
| Child Health | ~12.5% | 9% | Under-sampled |
| Mental Health | ~12.5% | 14% | Adequate |
| Population Health and Ethics | ~12.5% | 11% | Adequate |
In this example the candidate has let the adaptive feed and personal comfort pull them into medicine while women's health and child health — a combined quarter of the paper — are neglected. No dashboard percentage would reveal this; only the distribution does.
The readiness signal: five conditions before you trust a number
A credible readiness measurement has to satisfy all five conditions, or it is not a signal:
- Unseen. The items must be ones you have not attempted, so memory cannot flatter you.
- Timed. Roughly 84 seconds per item, to expose pacing failures under load.
- Mixed. Blueprint-proportioned across all six disciplines, not a single-topic drill.
- No assistance. No explanations, notes, tutor or search open during the block.
- Adequate sample. Enough items — a full 150 where possible, at least 40–50 for an interim read — that one lucky run does not dominate.
MedLumen's "timed, mixed" mocks can meet the first four conditions if you hold them back as assessment assets rather than burning them during casual study. The moment you have already seen the mock items, they stop being unseen and their score loses meaning.
Override rules: what to force into your queue
An adaptive feed optimises for your recent errors, not for the exam's low-frequency, high-consequence content. Manually override it to guarantee exposure to material the algorithm will under-serve:
- Low-volume disciplines you happen to score well on early (often population health and ethics), so a small strong sample does not hide a thin base.
- Image-dependent items — rashes, ECGs, radiographs, fundoscopy — which candidates who study on a phone skate past.
- Ethics, consent, mandatory reporting and Australian medico-legal scenarios, which reward jurisdiction-specific knowledge, not clinical recall.
- Calculations and safe prescribing grounded in Australian references — the Australian Medicines Handbook and Therapeutic Guidelines (eTG) — rather than any overseas formulary.
- Immunisation and screening aligned to the National Immunisation Program and Australian screening programmes, which differ materially from UK and US schedules.
Worked example: turning a dashboard into next week's quotas
Suppose your MedLumen dashboard (hypothetical, for illustration only) shows: first-attempt accuracy 61% overall; medicine 68%, surgery 58%, women's health 49%, child health 52%, mental health 66%, population health 71%; average 96 seconds per item; and the attempted distribution above. Translate that into quotas rather than a prediction:
- Women's health and child health are both below 55% first-attempt and under-sampled — set the largest quota here: 40 fresh items each this week, drilled to an error-code review.
- Surgery is near target share but under 60% — 30 fresh items.
- Pacing at 96 seconds is over budget — every block runs strictly timed at 84 seconds per item from now on.
- Medicine, strong and over-sampled, gets a maintenance quota only: 15 mixed items.
- One unseen, mixed, timed 50-item block at week's end, held back and scored once, as the actual readiness read.
Notice there is no "you will pass" statement. The dashboard sets where the work goes, not what the result will be. Predicting a pass from a training percentage is exactly the error this audit exists to prevent.
A seven-day pattern for international graduates
This loop gives MedLumen one clearly defined job — content review and weak-area drilling with Australian-context explanations — and gives iatroX a different job: unseen, mixed, timed transfer measurement. No claim is made about either platform's internal algorithms.
| Day | MedLumen (learning job) | iatroX (measurement job) |
|---|---|---|
| Mon | 40 items in the two weakest disciplines; error-code each miss | — |
| Tue | Read eTG/AMH notes on Monday's misses; 20 transfer items | — |
| Wed | 40 items in the next weakest discipline | Unseen mixed 20-item timed block; log first-attempt only |
| Thu | Image and ECG set; ethics/medico-legal set | — |
| Fri | 30 mixed items; re-test Monday's error codes | — |
| Sat | Light review of flagged explanations | Unseen mixed 50-item timed block, scored once |
| Sun | Rest or read specifications; plan next week's quotas from Saturday's block | — |
The iatroX blocks are unseen relative to your MedLumen practice, which is what makes them a fair transfer test rather than a rehearsal.
Decision checklist: continue, supplement, switch or stop
Base the call on measurable gaps, never on novelty or on how much you have already spent:
- Continue MedLumen if first-attempt accuracy is rising across disciplines and its Australian-context explanations are teaching you management you did not know.
- Supplement it if your attempted distribution is skewed or your unseen mixed scores lag your in-bank percentage by more than a few points — add a second, independent measurement bank.
- Switch the primary bank if repeated audits show thin coverage of a whole discipline, or if explanations lean on non-Australian guidance you keep having to correct.
- Stop adding new questions and move to pure mixed mocks once every discipline clears its floor on unseen items and pacing is under budget — more volume then adds fatigue, not readiness.
The bottom line
MedLumen is a reasonable, Australian-facing AMC MCQ bank with analytics that are genuinely useful if you read them as training telemetry rather than a score. Its adaptive feed is a study aid, not a model of the exam, and its headline percentage is the least trustworthy number on the screen. Elevate first-attempt accuracy on unseen, mixed, timed blocks; audit your attempted distribution against the AMC weighting; and force in the low-frequency, image-heavy, jurisdiction-specific content the feed will otherwise starve. Do that and the dashboard earns its place.
Frequently asked questions
Is MedLumen enough for AMC MCQ on its own? For most candidates, one bank is a foundation rather than a complete preparation. MedLumen's vendor-reported "732+" items (as at 19 July 2026) can build and drill knowledge, but a single bank cannot give you an independent, unseen measurement of readiness — once you have seen its questions, its scores flatter you. Pair it with a second, blueprint-proportioned source you hold back for mixed timed assessment, and anchor your clinical facts on Australian references.
Which AMC MCQ component does MedLumen not reproduce well? MedLumen prepares the written single-best-answer paper only; it does not reproduce the AMC Clinical Examination, the observed-consultation stage that follows the MCQ on the standard pathway. It also cannot substitute for the exam's fixed-length, whole-blueprint sampling: an adaptive weak-area feed is the opposite of the balanced paper you will actually sit, so its day-to-day percentage is not a fidelity match for exam conditions.
How many MedLumen questions should I complete per day for AMC MCQ? There is no single correct number; quality of review beats raw volume. A sustainable pattern for a working doctor is roughly 40–60 fresh items per study day, each with a coded error review, plus periodic held-back mixed blocks — enough to move first-attempt accuracy without so much throughput that you stop reviewing misses properly. Scale to your weeks-to-exam and protect time for the Australian-guideline reading behind each miss.
When should I stop using MedLumen and move to mixed mocks? Stop drilling new topic-filtered questions once every discipline clears a first-attempt floor on unseen items and your pacing sits under about 84 seconds per item. At that point additional single-topic volume mostly adds fatigue; your marginal gains come from full-length, mixed, timed mocks that rehearse endurance and decision-making across the whole blueprint under realistic pressure.
How should I combine MedLumen with iatroX without duplicating practice? Give each a distinct job. Use MedLumen for content learning and weak-area drilling, and use iatroX purely for unseen, mixed, timed measurement — never load the same items into both. Because iatroX questions are independent of your MedLumen history, an iatroX block functions as a fair transfer test of whether MedLumen's learning is sticking, rather than a rehearsal of questions you have already memorised.
Editorial notes and references
Written by Dr Kolawole Tytler, NHS GP and founder of iatroX. Last checked 19 July 2026; all MedLumen figures are vendor-reported from its product page on that date and may have changed since — verify the current count, price, currency and features on the live product page before relying on them.
Disclosure: iatroX operates its own AMC MCQ question bank and Socratic Tutor and is therefore a competitor to MedLumen. This audit deliberately confines iatroX's role to the unseen, mixed, timed measurement job that MedLumen's adaptive feed is not designed to serve, and makes no claim about MedLumen's or iatroX's internal algorithms. Corrections are welcome via the feedback route on iatrox.com.
References: AMC Computer Adaptive Test (CAT) MCQ Examination and AMC MCQ Specifications (V8, 9 September 2025), amc.org.au; MedLumen AMC CAT MCQ product page, medlumen.io; Therapeutic Guidelines (eTG), the Australian Medicines Handbook and the National Immunisation Program for Australian management standards. Internal reading: Your Q-Bank Percentage Is Not Your Exam Score, the completion-is-not-coverage blueprint matrix and the iatroX comparison hub.