This workflow is for physician associates preparing for the PA Registration Assessment (PARA) who already practise on Geeky Medics' AI OSCE stations and want the automated scoring to mean something. It addresses the clinical-simulation layer and its automarked feedback, not the written paper. The principal limitation is straightforward: that feedback is a checklist proxy, not the official Royal College of Physicians examiner judgement, and Geeky Medics carries no PARA-specific written bank—so treat every score as a training signal, never a verdict.
What Geeky Medics offers for PARA right now
Geeky Medics is a clinical-skills and OSCE platform rather than a PARA product, so read the table below as a map of what you can borrow, not a claim that the tool is built for this exam. Figures are vendor-reported and last checked 20 July 2026; confirm the live numbers and the current price before you rely on them.
| Feature | What we found (20 July 2026, vendor-reported) |
|---|---|
| OSCE stations | 1,300+ across the platform |
| AI virtual patients | 800+ scenarios (history taking, counselling, examination); typed or voice input |
| Automated feedback | Automarking of examiner checklists; a virtual examiner flags strengths and areas to improve |
| Written question content | MLA AKT, PSA, ECG/ABG/blood-test/radiology case banks; flashcards; notes |
| PARA-specific product | None advertised; content is general UK clinical-skills and UKMLA-aligned |
| Price | Not published on the AI OSCE page; free tier plus a paid subscription/bundle—verify the current price on geekymedics.com |
The honest finding comes first. Geeky Medics is genuinely strong at the one thing most question banks cannot give you: a realistic consultation to rehearse against, with instant automarking. That is the OSCE-facing half of PARA, and it is worth including in the revision stack. But the automarking runs against generic OSCE checklists, not the official PARA station rubric, and there is no dedicated PARA written bank here. iatroX sits on the other side of the exam—it supplies the applied-knowledge and single-best-answer layer plus unseen measurement, and it does not replace an OSCE simulator. Used together, each covers what the other does not claim to do.
PARA: two components, two different skills
The PA Registration Assessment is delivered by the Royal College of Physicians' Faculty of Physician Associates, and physician associates have been GMC-regulated since December 2024. It has two components: a written applied-knowledge examination built from single-best-answer questions, and an OSCE. Verify the current written question count, the OSCE station count and the pass standard on rcp.ac.uk, and anchor your revision to the GMC PARA content map rather than to any third-party syllabus. The distinction shapes this whole workflow: Geeky Medics' automated stations rehearse OSCE-style skills, while the written paper needs a separate SBA measurement layer. iatroX covers the written layer; it is not an OSCE tool, and this article never asks it to be one.
Build a representative case matrix
Before you can calibrate feedback, you need cases that spread across the blueprint rather than clustering on your comfort zone. Build a small matrix that varies domain, acuity, patient age, clinical complexity and the communication challenge, so that a good score on one axis cannot hide a weakness on another.
| Case | Domain | Acuity | Patient age | Complexity | Communication challenge |
|---|---|---|---|---|---|
| 1 | Cardiorespiratory | Acute | 68 | Multimorbidity | Breathless and anxious |
| 2 | Abdominal | Subacute | 34 | Single system | Guarded, minimal history |
| 3 | Neurology | Acute | 52 | Diagnostic uncertainty | Aphasia, collateral needed |
| 4 | Mental health | Non-acute | 23 | Risk assessment | Distressed, safeguarding |
| 5 | Endocrine/metabolic | Subacute | 45 | Investigation-led | Health beliefs, adherence |
| 6 | Older-adult/frailty | Acute | 81 | Polypharmacy | Sensory impairment, capacity |
Six cases is enough to expose a recurrent pattern without turning the week into an audit exercise. Keep the matrix; you will reuse it for deliberate variation later.
Record the first attempt as the baseline
Run each case once, end to end, without pausing, restarting or peeking at the mark scheme. Do not read the automated feedback until you have finished. The first, uninterrupted attempt is your baseline—the closest thing you have to how you will perform under real time pressure. If you stop to correct yourself mid-station, you have measured your editing, not your consultation. Save the transcript or the checklist output so you can compare it against a human-reviewed standard afterwards.
Score twice, and record every disagreement
Now score the same attempt two ways. First, take the platform's automarked result at face value. Second, mark the identical transcript yourself against the official PARA domains—or, better, have a supervising clinician or study partner mark it—using the RCP station descriptors as the reference. Write both scores side by side and, crucially, log every disagreement: places where the automarker credited a tick you would not have given, or penalised something a real examiner would have accepted. Those disagreements are the calibration data. Over six cases you will start to see the machine's systematic biases—typically over-crediting fluent phrasing and under-crediting safety-netting, capacity or ideas-concerns-expectations that were implied rather than spoken in the exact expected words.
Turn feedback into two observable behaviours
The failure mode here is drowning in a twenty-point improvement list. Instead, convert each case's feedback into exactly two observable behaviours to change on the next case—things a watching examiner could tick or not tick. "Communicate better" is not observable. "Offer an explicit safety-net with a named timeframe and a red-flag symptom" is. "Confirm the patient's main concern in their own words before I close" is. Two behaviours per case, carried forward and checked, will move a score further than a page of generic advice you will never operationalise.
Use deliberate variation, not the same script
Recognition is the enemy of transfer. Do not re-run the same station until the score climbs; that only teaches you the mark scheme. Instead, keep the clinical principle and change the surface: the same safety-netting skill against a different patient agenda, the same diagnostic reasoning with an added comorbidity, the same explanation under tighter time pressure. If you rehearsed chest pain in a calm 60-year-old, next take it to an anxious 40-year-old with a family history and a hidden agenda. You are training the underlying competency, not memorising a performance.
Exit standard
You are ready to move on from calibration when three things hold: your performance is consistent across unseen cases rather than only on ones you have repeated; your own marking and the platform's marking now agree most of the time, so you can trust the automated signal within a known margin; and there is no recurrent safety-critical omission—no domain where you keep missing the same red flag, capacity check or escalation. Until those hold, keep varying and re-scoring rather than adding volume.
A seven-day plan: one job each for Geeky Medics and iatroX
Give each tool a single, defined job. Geeky Medics runs the consultation and its automarking; iatroX measures the written applied-knowledge layer on unseen SBAs. No proprietary-algorithm claims are needed—the loop is just watch, test, space the misses, and measure on fresh items.
- Day 1: Run cases 1–2 from the matrix cold; save baselines. Score twice; log disagreements.
- Day 2: Convert Day 1 feedback into two behaviours each; re-run the same principle with a varied patient agenda.
- Day 3: Sit a fresh, timed 30-item PARA-level SBA block in iatroX on the written domains; review misses by error type.
- Day 4: Run cases 3–4 cold; score twice; carry two behaviours forward.
- Day 5: Deliberate variation on your two weakest stations; add a mental-health/safeguarding case.
- Day 6: Second unseen iatroX SBA block on a different domain mix; compare error types, not wording, against the station feedback.
- Day 7: Run cases 5–6 as a mini-mock; tally calibration agreement and any recurrent safety omission; set next week's two priorities.
Decision checklist: continue, supplement, switch or stop
- Continue with Geeky Medics if your automarker-versus-human agreement is improving and the stations are still surfacing new communication or examination gaps.
- Supplement (the usual answer) if your consultation skills are progressing but your written applied-knowledge scores on unseen SBAs are flat—that is iatroX's job, not the simulator's.
- Switch the automated feedback for human marking on any domain where the machine and a clinician keep disagreeing; trust the human on capacity, safeguarding and safety-netting.
- Stop adding station volume once performance is consistent on unseen cases with no recurrent safety omission; more repetition then buys recognition, not readiness.
Bottom line
Geeky Medics gives physician associates a rare asset—a rehearsable consultation with instant feedback—but the score is a checklist proxy, not the RCP rubric, and there is no PARA written bank inside it. Calibrate the automated feedback against the official domains, act on two observable behaviours at a time, and use iatroX to measure the applied-knowledge layer on unseen questions. That division of labour is honest about what each tool can and cannot do.
Frequently asked questions
Is Geeky Medics enough for PARA on its own? No. It is a strong option for rehearsing the OSCE-style consultation and getting fast automated feedback, but it does not carry a PARA-specific written applied-knowledge bank, and its automarking is not the official RCP examiner standard. Treat it as the simulation layer and pair it with a written SBA bank and, ideally, human marking on the domains that matter most for safety.
Which PARA component does Geeky Medics not reproduce well? The written applied-knowledge examination. Geeky Medics' strength is the OSCE-facing simulation; it does not advertise a PARA-mapped single-best-answer paper, so the knowledge component is where you will need a separate, blueprint-mapped bank such as iatroX and the GMC PARA content map to guide coverage.
How many unseen Geeky Medics cases or stations should I preserve for final PARA calibration? Keep a small, deliberately unseen set—around four to six varied stations you have never marked—to run as a mini-mock in the final fortnight. Their value is that they are uncontaminated by rehearsal, so they give you a cleaner readiness signal than any station you have already drilled.
When should I stop using Geeky Medics and move to mixed mocks? When your performance is consistent on unseen stations, your automarker-versus-human agreement is stable, and no domain shows a recurrent safety-critical omission. At that point, extra single-station repetition mostly rehearses familiarity; a full mixed mock under time pressure becomes the more informative test.
How should I combine Geeky Medics with iatroX without duplicating practice? Give them non-overlapping jobs. Geeky Medics runs the consultation and its automated feedback; iatroX measures the written applied-knowledge layer on fresh, unseen SBAs. After each iatroX block, compare the error type—reasoning, knowledge gap, or safety omission—against your station feedback, rather than re-answering the same items in both tools.
Editorial notes and references
Written by Dr Kolawole Tytler, NHS GP and founder of iatroX. Last checked 20 July 2026. Platform figures (question and station counts, features, pricing) are vendor-reported and change without notice—verify the live details on the vendor's site before relying on them. Disclosure: iatroX operates a UK question bank that competes with parts of the tools discussed here; this article confines iatroX's role to the written applied-knowledge and unseen-measurement layer that Geeky Medics does not claim to provide, and it does not position iatroX as an OSCE simulator. Corrections are welcome via the feedback route on iatrox.com.
References: Royal College of Physicians / Faculty of Physician Associates PARA regulations and the GMC PARA content map (rcp.ac.uk; gmc-uk.org); Geeky Medics AI OSCE practice and platform pages (geekymedics.com); iatroX, Your Q-Bank Percentage Is Not Your Exam Score; iatroX, calibrating automated OSCE and SAQ feedback; the iatroX comparison hub and the PARA 2026 format guide.
Try an unseen PARA question in iatroX and compare the error type →