A question bank is the right tool for one job in the Physician Associate Registration Assessment (PARA): verifying that you can select the correct answer, quickly, across the knowledge blueprint. It is the wrong tool — in fact, no tool at all — for the things the exam also tests: performing a station, taking a history from a real person, examining, communicating, carrying out core procedures, and doing all of it against a marking rubric you never see as a multiple-choice option. If you understand the PARA exam format, you already know why: half the assessment is an OSCE, and an OSCE cannot be sat with a mouse. This article names precisely what MCQ practice cannot reach and gives you a way to train each part.
This piece is the modality-gap companion to the PARA content-gap checklist; read that for coverage verification, and this for the performance skills a bank leaves untouched.
The skills ordinary MCQ practice cannot assess
State it plainly. A PARA MCQ bank cannot assess: your ability to gather a history through open and targeted questioning; your examination technique and the physical findings you elicit; your communication and shared decision-making with a patient or relative; your core procedural skills; your management delivered as a plan, sequenced and safety-netted, rather than picked from five options; your behaviour under observation in a timed station; and whether you are revising from the current version of the GMC blueprint at all. Everything a bank measures is downstream of one act — recognising the best option — and the OSCE is built specifically to test the acts that come before and after that recognition.
Official format map
PARA is delivered by the RCP for the GMC (which began regulating physician associates on 13 December 2024). It has two components, and the split is the whole point of this article (verified against RCP/GMC, last checked 20 July 2026):
- Knowledge-Based Assessment (KBA): 200 single-best-answer questions online, across four one-hour papers of 50. This is the layer a bank trains.
- OSCE: 16 stations — 14 scored plus 2 rest — each scored station 8 minutes (7-minute warning) with 2 minutes reading, 35 marks per station (490 total). To pass you must reach the standard and pass a minimum of 9 stations. The OSCE is marked across four domains: Communication 30%, Diagnosis 30%, Clinical management 25%, and Therapeutic management/core procedural skills 15%, with at least one station from each of nine mandatory specialties (acute care, cardiovascular, child health, gastrointestinal, mental health, musculoskeletal, neurosciences, obstetrics and gynaecology, respiratory).
Both components map back to the GMC PARA content map (four domains; 18 areas of clinical practice). A bank can cover the content map's knowledge; only live practice covers its skills and procedures.
Knowledge versus performance: what a correct answer proves
A correct single-best-answer selection proves something narrow and real: that, given a curated stem and five options, you can recognise the best one. That is genuine knowledge and it matters. But notice what it does not prove. It does not prove you can obtain the information in the stem from an actual patient who does not volunteer it in tidy sentences. It does not prove you can perform the examination that generates the finding. It does not prove you can explain the diagnosis to a frightened relative, or sequence a management plan and safety-net it aloud, or maintain the interaction while a clock runs and an examiner marks a rubric. The stem hands you the hard part — the synthesis — for free. The OSCE takes it back. That gap between recognising and doing is the modality gap, and it is where confident bank-scorers lose OSCE stations.
The under-tested skills, broken down
OSCE and clinical skills
This is not one skill but a cluster: history-taking, focused examination, communication and shared decision-making, core procedures, and integrated management under time. Each is trained by doing it under observation and being marked against a rubric — never by reading. A bank can build the knowledge a station rests on (you cannot manage chest pain you do not understand), but it cannot build the performance. Treat the bank as the foundation and the station practice as the build.
Current GMC blueprint versioning
The second under-tested "skill" is really a discipline: knowing, and checking, which version of the blueprint you are training against. MCQ banks — especially third-party ones — can lag the official content map. Candidates absorb an outdated emphasis without noticing, because the bank never tells you its blueprint is stale. The skill is to go to the source (the GMC content map and RCP KBA/OSCE blueprints), confirm the dated version in force for your diet, and audit your bank against it rather than trusting it to be current.
For each skill: behaviour, task, feedback, exit standard
| Skill | Observable behaviour | Deliberate-practice task | Feedback source | Exit standard |
|---|---|---|---|---|
| History-taking | Elicits the key positives and negatives through structured questioning within the station time | Timed history stations with a simulated patient who withholds detail | Clinician/examiner against OSCE rubric | Reaches a safe differential and plan within 8 minutes, unprompted |
| Examination | Performs a fluent, systematic focused examination and reports findings | Peer- or tutor-observed examination drills by system | Clinician observer | Correct technique and interpretation without prompting |
| Communication | Explains clearly, checks understanding, shares decisions, shows empathy | Breaking-news and explanation stations, recorded and reviewed | Simulated patient + clinician | Patient-rated clarity plus examiner-rated domain pass |
| Core procedures | Completes the procedure safely to protocol | Skills-lab repetition on manikins/models | Skills tutor sign-off | Independent, safe, protocol-adherent completion |
| Integrated management | Sequences and safety-nets a plan aloud under time | Full mixed OSCE circuits | Examiner against all four domains | Passes the domain and hits the minimum-station threshold |
| Blueprint currency | Revises only from the version in force | Quarterly source-check against GMC/RCP pages | The official pages themselves | Confirmed current version, logged with date |
A four-week modality ladder
Do not jump from reading to a full mock. Climb:
- Week 1 — isolated skill. Drill one component at a time: a history frame, one examination system, one procedure. Low stakes, high repetition, immediate correction. Your bank runs in parallel to keep the underlying knowledge warm.
- Week 2 — coached case. Combine skills into a whole station, but with a coach who can pause you, rewind and correct. The aim is a correct process, not yet a clean run.
- Week 3 — timed integrated case. Now add the clock and the rubric. Run stations at 8 minutes with 2 minutes reading, marked against the four OSCE domains. Log which domain costs you marks.
- Week 4 — unseen simulation. Full mixed circuits and, on the knowledge side, unseen timed KBA blocks you have not previously reviewed. This is the readiness signal: performance on fresh material under exam conditions.
When AI feedback helps, when it does not, and when you need a clinician
Automated feedback has a real place and a hard limit. It helps for structured, checkable things: whether your differential covered the likely diagnoses, whether you named the right investigations, whether a written management plan is complete, whether your KBA reasoning was sound. It is unreliable for anything requiring judgement of a live human interaction — rapport, the appropriateness of your empathy, whether an examination was actually performed competently, whether a procedure was safe — because those are observed behaviours a text model cannot see. And a clinician or examiner is required for the final calibration of any OSCE domain, for procedural sign-off, and for the "would this pass?" judgement that carries real marks. Before you trust any automated score, calibrate it against a human rubric — the method is in how to calibrate AI-graded SAQs and OSCEs. Use AI to rehearse volume and structure; use a clinician to certify performance.
A balanced case matrix
Left alone, candidates rehearse the same three comfortable stations. Force breadth with a matrix that crosses the nine mandatory OSCE specialties against station types so you cannot practise only familiar scenarios:
| Specialty \ Type | History | Examination | Communication | Procedure/Data |
|---|---|---|---|---|
| Acute & emergency care | ✔ | ✔ | ✔ | ✔ |
| Cardiovascular | ✔ | ✔ | · | ✔ |
| Child health | ✔ | · | ✔ | · |
| Gastrointestinal | ✔ | ✔ | · | ✔ |
| Mental health | ✔ | · | ✔ | · |
| Musculoskeletal | · | ✔ | · | ✔ |
| Neurosciences | ✔ | ✔ | · | ✔ |
| Obstetrics & gynaecology | ✔ | · | ✔ | · |
| Respiratory | ✔ | ✔ | · | ✔ |
Tick cells as you rehearse them; empty rows and columns are your real revision plan, not the stations you already enjoy.
Red flags that you are training the wrong thing
- Memorised scripts. If your history is a recitation rather than a response to the patient in front of you, an examiner will see it and the marks will follow the patient, not the script.
- Repeated cases. Running the same three stations to a polish trains recognition, not skill. Rotate.
- Generic feedback. "Good communication" is not feedback; a domain-referenced note ("lost management marks by not safety-netting") is.
- Uncalibrated scoring. A number from a tool you have never checked against a real rubric is decoration.
- No official-rubric check. If you have never marked a practice station against the RCP OSCE domains, you are guessing at the standard.
Worked example: one candidate, two layers
Priya sits her KBA mock at 74% first-attempt accuracy on unseen mixed blocks — a solid knowledge signal, and on that number alone she feels ready. Then she runs a single timed OSCE circuit and clears only 6 of the mandatory-specialty stations against a tutor's rubric: her knowledge is sound, but she talks over the simulated patient, forgets to safety-net, and spends her eight minutes explaining rather than examining. The two numbers tell different stories because they measure different modalities. Priya's plan writes itself from the gap: she keeps her knowledge warm with two short unseen blocks a week, and moves the bulk of her remaining time onto the ladder — isolated communication drills first, then coached stations, then timed circuits marked on the four OSCE domains. Three weeks later her station pass-rate has risen to 10 of 12, and her communication-domain marks — the ones the KBA never touched — have stopped costing her. Note what did not happen: she did not grind her KBA percentage from 74% to 80%, because that was never the failing. The modality gap was, and only station practice closed it. If you are reading your own results this way, the question is never "is my percentage high enough?" but "which layer is the weak one, and am I training that layer or the comfortable one?"
The bottom line
A question bank is necessary and not sufficient for PARA. It verifies that you can recognise the best answer across the knowledge blueprint, which is real and worth having — but PARA also runs a 16-station OSCE that tests the acts a multiple-choice item hands you for free: gathering, examining, communicating, performing and sequencing under a clock and a rubric. Treat the two as different modalities with different training methods. Keep your knowledge warm on the bank, prove it on unseen timed blocks, and spend the modality-specific time where the marks actually hide — on observed, rubric-marked station practice across all nine mandatory specialties. The candidate who does both walks in ready; the one who only does the comfortable half walks in surprised.
FAQ
How do I know whether I have covered the full Physician Associate Registration Assessment blueprint? You have covered it only when both layers are evidenced: the knowledge layer through a blueprint coverage table showing every content-map area sampled on unseen items, and the skills layer through rehearsed OSCE stations across all nine mandatory specialties, each marked against the RCP domains by a competent observer. A full-coverage claim that rests only on MCQ completion is, by definition, half a claim, because it ignores the OSCE. Verify the areas against the current GMC content map, since the version in force can change between diets.
Can one question bank be enough for Physician Associate Registration Assessment? No — a bank, however good, trains only the KBA, and PARA also examines an OSCE that no bank reproduces. Even within the knowledge component, a single bank becomes a recognition exercise once most items are seen, which is why a second unseen bank and dedicated OSCE practice are the missing pieces. Think in layers: one bank for coverage, unseen items for measurement, and live station practice for the modality a bank cannot touch.
What should I measure instead of my overall Q-bank percentage for Physician Associate Registration Assessment? For knowledge, measure first-attempt accuracy on unseen, timed, mixed blocks by content-map area, plus your rate of high-confidence errors; for the OSCE, measure domain-referenced pass/fail on rehearsed stations and how many stations you clear in a full circuit. The overall bank percentage says nothing about station performance and is inflated by review on the knowledge side, so it is close to the least informative number you own. Track the layer-specific signals instead.
When should I stop doing new Physician Associate Registration Assessment questions? Stop adding new knowledge questions when your unseen first-attempt accuracy is stable and adequate across every area and your errors are random rather than clustered — but recognise that reaching that point does not mean you are ready, because the OSCE clock may not have started in your preparation at all. In most cases the honest answer is to shift modality rather than stop entirely: move surplus knowledge-revision time into station practice, which is usually the under-trained half. Let the measured gap in each layer, not question fatigue, drive the switch.
Which Physician Associate Registration Assessment resource should I use for my weakest component? If your weakest component is the OSCE or any clinical skill, use a structured, observed OSCE course — that is the only modality that trains it, and no MCQ product substitutes. If it is knowledge or interpretation, use a question bank strong in that domain, adding image, ECG and data-heavy sets for interpretation. iatroX supports the knowledge and unseen-measurement layer through the PARA bank on the quiz landing page and is deliberately not positioned as an OSCE simulator, so pair it with live clinical-skills practice whenever the weak component is a station.
Editorial notes and references
Written by Dr Kolawole Tytler, NHS GP and founder of iatroX. Last checked 20 July 2026. Exam-format figures come from RCP and GMC pages on the date shown and can change; verify the current OSCE station count, timing, marking domains and blueprint version on rcp.ac.uk and gmc-uk.org. Disclosure: iatroX operates a UK question bank that competes on the PARA knowledge layer; it is presented here only as the knowledge and unseen-measurement layer and is explicitly not an OSCE or clinical-skills simulator. Corrections are welcome via the feedback route on iatrox.com.
References: RCP PARA information for candidates and OSCE blueprint (rcp.ac.uk); GMC information about the PARA and the PARA content map (gmc-uk.org); the PARA content-gap checklist; how to calibrate AI-graded SAQs and OSCEs; Your Q-Bank Percentage Is Not Your Exam Score; the iatroX comparison hub.