Medicine offers few spectacles more exposing than the clinical problem solving conference: a single physician seated before an audience of trainees and colleagues, handed a mystery case one fragment at a time, and asked to reason aloud, in real time, toward a diagnosis that may never fully arrive. A new perspective article published on 26 August 2026 in the Journal of General Internal Medicine offers the first practical playbook for this high-stakes teaching role, built around a striking claim: success depends less on reaching the correct diagnosis than on delivering structured, transparent reasoning in the face of uncertainty. Gurpreet Dhaliwal of the San Francisco Veterans Affairs Medical Center and the University of California, San Francisco, together with Sarah Goglin of Zuckerberg San Francisco General Hospital and André N. Sofair of the Yale University School of Medicine, argue that when faculty take the discussant’s seat, they model a quintessential physician skill — the ability to think clearly, flexibly, and humanely when the answer is not yet known.

The clinical problem solving (CPS) exercise is a structured case presentation format in which a clinician analyzes a clinical scenario in real time. The case unfolds sequentially — history first, then vital signs, physical findings, laboratory data, and imaging — so that learners watch an experienced mind weigh each new data point and adjust its hypotheses accordingly. Because the discussant sees the case only as it is revealed, the exercise mirrors real clinical work: incomplete information, evolving data, and decisions that cannot wait. The format is widely deployed across teaching conferences, medical journals, and podcasts, and decades of scholarship have established reasoning and judgment as the core function of medical training. Learners consistently identify an attending physician’s articulated clinical reasoning as the most valuable component of clinical instruction. Yet while published resources exist to guide the presenter who narrates the case, the authors note, guidance for the discussant — the clinician who improvises the analysis live — has been conspicuously absent.

That gap matters, because faculty who serve as CPS discussants fulfill core clinician–educator milestones, including reflective practice and role modeling, while exposing trainees to a wider spectrum of diagnostic styles and approaches than any single teacher could provide. Some faculty, however, find the role daunting, deterred by public speaking anxiety or by the belief that they must reach the correct final diagnosis. Dhaliwal and colleagues push back against that assumption. Although clinical experience best equips a discussant to reason through a case and propose a logical diagnosis, they write, any faculty member can succeed in the role, particularly when the aim is transparent reasoning rather than diagnostic perfection. Their aim is for more teachers to accept invitations into the role, establishing CPS conferences as a central pillar of training programs — an arena where professional judgment, not artificial intelligence, remains the cornerstone of patient care.

Preparation, the guide emphasizes, begins long before the conference hall. The foundational steps are threefold: strengthening clinical analytical skills, observing how skilled discussants work through cases, and rehearsing in low-stakes settings that build comfort with speaking and thinking aloud. To build the clinical foundation, the authors recommend working in patient care settings that reward questioning and analysis, such as teaching roles; seeking cases at the edge of one’s comfort zone; and consulting colleagues on common and complex presentations. Reflective practice deepens this base: reviewing the charts of diagnostically challenging patients one has cared for and posing metacognitive questions — how would I explain my reasoning? — sharpens the insight required for both bedside care and public case analysis. The point is to convert everyday clinical work into a continuous, self-directed laboratory for diagnostic thinking.

Observation is the second pillar. Aspiring discussants are urged to study how experienced clinicians take cases apart in morning report, morbidity and mortality conferences, subspecialty rounds, and medical podcasts, attending closely to the architecture of those performances: how often the analyst reframes the case as new data arrive, how long they hold the floor when making a single teaching point, how they frame uncertainty without surrendering authority, and how they draw listeners into the chase. The authors advise noting what makes a compelling analysis compelling, and cataloguing the strategies seasoned clinicians deploy when they venture into unfamiliar territory. Reading published CPS articles exposes the way spoken reasoning is edited and structured for the page, a useful template for organizing the same ideas aloud. The exercise is apprenticeship by proxy — learning the choreography of thinking out loud before performing it.

Rehearsal opportunities are more accessible than they might seem. Facilitating any case conference builds public speaking stamina without the exposure of the hot seat, because the role demands the same core competencies: summarizing clinical information, responding to evolving data, and clarifying complex points. Multi-discussant CPS formats, in which several clinicians share the analytical load for a single case, and CPS-style journal clubs, where the focus alternates between a discussant and the group, cultivate spontaneous clinical commentary in a small and supportive setting. One-on-one sessions with a mentor provide immediate feedback, while small student conferences and virtual platforms, where the audience is not directly visible, let clinicians refine reasoning aloud, pacing, and the framing of uncertainty. Abridged solo formats, such as a clinical pathologic conference with the full case provided in advance, balance structured commentary with spontaneous discussion. The authors even suggest having a large language model build a practice session: divide any case into segments, present each in turn, solicit the trainee’s analysis, and return feedback.

During the conference itself, the discussant’s task is to guide the audience on an educational, engaging, and authentic diagnostic journey. The guide recommends opening by setting the stage: briefly explaining that the goal is to model reasoning, including the insights and struggles ahead, and, when one exists, introducing a personal method for dissecting complex cases, such as organizing the problem by host factors, epidemiology, tempo, and syndrome. As the case unfolds, the discussant should verbalize how information is grouped and prioritized — why the fever is being highlighted while the well-controlled diabetes is set aside — pausing periodically to summarize and to articulate a problem representation, the concise synthesis of findings into a recognizable pattern, or to sketch a problem list on a whiteboard. Diagnostic schemas deserve explicit airtime: rather than reciting an exhaustive differential, the discussant should share the framework through which the problem is approached — the general categories of thrombocytopenia, say — beginning as a generalist would even when a specialist’s vantage point beckons, then shifting from broad frameworks to leading diagnostic candidates and the justifications that rank them, signposting each move so the audience sees the structure of clinical reasoning.

Realism is a recurring theme. The discussant should think as they would in the actual setting where the case took place, such as an urgent care clinic, and while it is fair to observe that rare conditions are disproportionately represented in teaching conferences, common diagnoses must be given their due first: if the suspected answer is sarcoid cardiomyopathy, the audience should hear why coronary artery disease or hypertension is not responsible for the heart failure before the exotic candidate is unveiled. Pacing matters as much as content. Reflections of roughly three to four minutes per case segment preserve the presenter’s momentum, and if new information does not modify the differential, the discussant should explain why and move on. When the data become bewildering, the guide sanctions honest recovery strategies: returning to notes, sketching a Venn diagram on the whiteboard, pausing to think, or asking the audience for ideas not yet considered. Either way, the discussant is obligated to commit to a leading hypothesis and explain why it outranks the alternatives, creating the reference point for the session’s signature finale: the cognitive autopsy.

The cognitive autopsy, whether the final diagnosis proved correct, incorrect, or somewhere in between, walks the crowd through one’s own reasoning and dissects the moments of success and failure — an overemphasis on a normal C-reactive protein, say, or a memory lapse about whether eosinophilic pneumonia typically produces peripheral eosinophilia. That retrospective analysis models humility and fallibility while preserving credibility, a balance the authors identify as the mark of a skilled discussant. Self-regulation sustains the performance: seek clarification rather than reason from a misheard data point, as in asking the presenter to display an electrocardiogram again; asking whether a blood smear was done is acceptable, but pressing or faulting the presenter for its absence is not, since hindsight is twenty-twenty and any presentation inevitably simplifies the complexity the original clinicians faced in real time. A private notepad serves as a memory aid, while a whiteboard can double as a shared canvas for causal diagrams and diagnostic schemas. Audience participation, meanwhile, should wait until the case has grown complex enough for outside input to help.

Reflection closes the loop. The period immediately after the conference should be used to archive case details, flag clinical topics that require further study, and probe how the reasoning might have been explained more effectively, with targeted feedback from the presenter or a trusted listener — did the hemolysis discussion confuse everyone, and did the final diagnosis sound committed or hesitant? The next invitation, the authors advise, should come before too much time passes. Then the guide turns to the question shadowing every modern classroom. Large language models can generate fluent, coherent case analyses and mimic the language of reasoning, but their outputs do not arise from genuine clinical thought: such systems hold no causal models of disease or illness, cannot integrate real-world constraints or patient values, and carry no sense of consequence or professional responsibility. Learners have access to all the medical knowledge in the world, the authors write, but they need their teachers to reveal the how and the why. There is no hierarchy of settings for making reasoning visible, and the gains transfer to one-on-one explanations with learners, patients, and colleagues. When teachers step into the discussant’s role, they enact the very capabilities machines lack, modeling the ability to think and communicate clearly, flexibly, and humanely in the face of uncertainty.