A virtual patient session ends where its educational value begins: with a list of things that went wrong, the question not asked, the red flag missed, the management step fumbled, and most students close the tab there, converting an exposure into an anecdote. The loop this article publishes converts it into learning instead, in five steps across two kinds of platform, because no single product should pretend to do everything: the simulation layer exposes gaps in performance, the retrieval layer repairs and retains the knowledge underneath them, and the join between the two is a workflow, not a feature.
Step one: run the consultation, capture the misses
Complete the virtual case, Geeky Medics-style AI patients, SimConverse scenarios or an institutional simulation, and before reading any feedback, write your own debrief in three lines: what you think you missed, what you were unsure of mid-consultation, and what you would do differently. Then read the platform's feedback against your list. The self-debrief matters for the same generation-effect reasons as attempt-first questioning: your own error hypothesis, corrected, encodes deeper than a scorecard absorbed, and the divergence between your list and the platform's is itself diagnostic, misses you predicted are knowledge gaps, misses that surprised you are awareness gaps, and the two repair differently.
Step two: convert performance errors into knowledge gaps
The translation step most workflows skip. A consultation error is a behaviour, forgot to ask about anticoagulants, but underneath it sits a nameable knowledge object, the bleeding-risk assessment for this presentation, and the conversion rule is to write each miss as a question you should have been able to answer: what makes this chest pain high-risk, what are the red flags in this headache, which drugs interact with the one this patient mentioned. Three to six such questions per case is typical, and cases generating none were rehearsal rather than diagnosis, valuable differently. This list, not the score, is the case's output.
Steps three and four: targeted repair, then spaced proof
Take the question list into the retrieval layer. Targeted practice: attempt blueprint-bank questions on each named topic, attempt-first as always, and where an answer is wrong for reasons you cannot articulate, run the Tutor pass, the Socratic interrogation of your reasoning that locates the actual misconception rather than re-explaining the topic, which is the iatroX configuration of this step and the rubric any tutoring layer should meet. Then schedule the proof: the topics enter spaced repetition for unaided return days later, because the loop's claim is retention, not a good evening, and only the delayed unassisted retest can cash it. The outcome-ladder logic runs through this joint: the consultation tested application, the repair works at knowledge, and the spaced retest verifies the rung that examinations pay.
Step five: rerun the case, perturbed
Close the loop where it opened, with the crucial variation: not the identical scenario, whose repetition teaches the script, but a perturbed one, different age, new comorbidity, altered emotional state, a complication added, so that the repaired knowledge must transfer rather than replay. Platforms differ in how genuinely their repeats vary, the unlimited-repetition question worth asking of any simulation product, and where variation is thin, perturb manually by choosing an adjacent case. A pass on the perturbed rerun is the only performance evidence the loop accepts; a pass on the identical rerun is memory wearing competence's costume. Run weekly, one or two cases fully looped beats five cases merely completed, and across a rotation the accumulated question lists become a personalised map of your actual clinical weak points, which no generic revision plan can supply.
Frequently asked questions
How long does one full loop take?
Roughly ninety minutes spread across a week: consultation and debrief thirty, conversion ten, targeted repair thirty, spaced retest built into normal question practice, perturbed rerun twenty; the spreading is the design, not a compromise.
Does the loop work for real placement encounters too?
Directly, with the consultation replaced by the ward round or clinic case and the platform feedback replaced by your supervisor's; the conversion, repair and retest machinery neither knows nor cares where the misses came from.
Which step do students most often skip, and what does it cost?
Step five, and it costs the transfer evidence: without the perturbed rerun, the loop proves knowledge repair but never performance change, which is the difference between feeling readier and being readier.
How should the loop feed OSCE-season planning?
The accumulated question lists cluster: recurring themes across cases, risk assessment, safety-netting, drug histories, are your genuine station weaknesses, and OSCE revision built from that map beats generic station lists by construction.
Can the loop run on group cases?
Well: shared consultation, individual debriefs and question lists, pooled repair topics, separate spaced retests; the group supplies perspective on the misses while the retrieval stays individual, which is the correct division.