How Clinical Judgment Is Actually Built — and Why So Little of It Can Be Taught

A resident can recite the entire taxonomy of cognitive bias from memory and still anchor on the wrong diagnosis under pressure, the same way a first-year would. A specialist with twenty years behind them can know less about a routine case outside their field than the intern shadowing them. A professor can run the clearest teaching round of the week and still fail to pass on the one thing that makes them good at the job. And the clinician who reads a deteriorating patient before the monitor does often cannot tell you how they knew.

Four failures. Four different research literatures, four different departments, four different fixes proposed over the years, none of which cite each other. Look closely and they are not four failures. They are one fact about clinical knowledge, showing up in four places that don't talk to each other.

None of these clinicians did anything wrong. They ran into the same wall from four different directions.

The assumption underneath medical training

Most medical education treats clinical knowledge as content: something that can be stated, written down, and handed from one person to another through teaching. This is true for a great deal of what a clinician needs. Dosing tables, diagnostic criteria, treatment algorithms: read it, teach it, test it, retain it. The model works.

The philosopher Michael Polanyi drew a distinction in the 1960s that clinical education has been slow to absorb. Alongside explicit knowledge, the kind that transfers through telling, sits tacit knowledge: understanding built through direct experience, compressed by repetition into something a person can use but not fully state. "We know more than we can tell," Polanyi wrote, and he meant it as a structural claim, not a comment about modesty.

Clinical judgment, the part of medicine that separates a competent clinician from a merely knowledgeable one, sits mostly on the tacit side of that line. It builds the way tacit knowledge builds: through exposure, not through being told. Training systems that treat it as explicit content, the way they correctly treat dosing tables, are applying the right model to the wrong kind of knowledge. What that mismatch actually looks like has been documented independently in four places.

Four places where that assumption breaks down

When clinicians try to catch their own bias

Anchoring, availability, premature closure: these have sat in medical curricula for decades, and diagnostic error rates have not meaningfully moved. The problem is not forgetting the material. Biases operate in System 1, the fast, automatic mode of thinking Daniel Kahneman described; awareness of them is a System 2 activity, slow and only partly available under real clinical load.

By the time a clinician consciously checks for anchoring, the anchored judgment has already shaped what evidence they noticed and what they discounted. One emergency physician's public account of their worst diagnostic miss makes the timing problem concrete: EMS mentioned a patient seemed to improve slightly after naloxone en route, and that single detail set the frame for everything after, overdose, then possible anoxic injury, with a stroke considered and talked out of because the patient was moving all four limbs too strongly.

It was a massive basilar artery stroke; the improvement had been coincidence. As the physician put it afterward, the department had also just been downstaffed from triple to double weekend coverage, and a second cardiac arrest was called mid-workup, a detail worth naming since Croskerry's research found structural fixes to clinical environments do more here than individual awareness training ever does.

The picture gets more specific under closer study. Norman and colleagues, reviewing the literature on diagnostic error, found that a substantial share of mistakes attributed to bias are better explained by knowledge deficits: the clinician simply didn't have the right illness script available. A clinician who looks like they're anchoring on the wrong diagnosis may be anchoring correctly on the only diagnosis their pattern library offers. A nurse who witnessed a near-identical case elsewhere, a patient waking from a procedure not making eye contact, speaking incoherently, thrashing all limbs, named the deeper issue plainly: a stroke with obvious one-sided weakness reads as a stroke immediately, but an undifferentiated, agitated patient with no clear history reads as almost anything else first. The trap isn't a failure of vigilance, it's that the correct diagnosis is wearing the costume of a dozen more common ones. The full mechanics of why awareness fails to protect against bias, and what the evidence says actually helps, are laid out in why cognitive bias persists in clinical decision making.

Not every diagnostic miss belongs in this bias framework. A physician who ordered a sinus CT for a headache patient, only to find days later that the same scan simply hadn't captured enough of the brain to show what turned out to be a large bleed, wasn't anchored on the wrong diagnosis, they were working within the honest limits of the imaging they'd chosen. A stutter that turned out to be moyamoya disease, caught only incidentally on a follow-up scan, sits in the same category: a presentation so rare that multiple clinicians discussing the case agreed no standard workup would reasonably have caught it sooner. These are knowledge-deficit misses, the kind Norman's research distinguishes from bias, not anchoring in a different costume.

As specialization narrows what stays sharp

Clinical knowledge does not simply accumulate across a career. Unused areas recede, quietly at first, in a pattern that echoes the shape of Ebbinghaus's forgetting curve without being the same phenomenon. Ebbinghaus studied memorized nonsense syllables in a laboratory, not clinical skill. What the medical-education-specific evidence actually shows comes from Custers's review of twenty empirical studies on knowledge retention: roughly two-thirds to three-quarters of medical knowledge survives the first year unused, dropping toward half in the years after. What decays fastest isn't the surface layer of facts, which can be looked up. It's the tacit layer, the calibrated judgment that only builds through repeated exposure, and why clinical knowledge decays, and what that means for expertise is easy to underestimate until you see it named specifically. An OB-GYN, responding publicly to a nurse practitioner asking whether to treat outside her specialty, laid out what that layer holds for something as ordinary as asthma: not the diagnosis itself, but which nebulizer solution to order, which subtype she's looking at, when a case has quietly stopped responding. None of that shows up on a label that just says "asthma." It's the part that erodes first while the label stays intact.

Years of experience and breadth of knowledge are not the same thing, and they sometimes work against each other. A highly specialized clinician can have a narrower working knowledge base than a resident several years into training, outside their own specialty. The same physician sharpened this with a second example: roughly 95 percent of the time, a pregnant patient with elevated liver enzymes, low platelets, and malaise has preeclampsia. The other five percent, acute fatty liver, a thrombotic microangiopathy, lupus, sepsis, depends on a volume of exposure a specialist who no longer sees the condition regularly has usually stopped accumulating.

When senior doctors try to teach it

A senior clinician's knowledge sits in two layers. The explicit layer, guidelines, criteria, named frameworks, transfers cleanly through teaching. The tacit layer, calibration built through thousands of encounters, does not, because it has compressed past the point where the person holding it can describe the process that produced it. Bereiter and Scardamalia named this split decades ago as formal versus embedded knowledge; Michael Eraut's later research on workplace learning builds on it directly, calling the embedded layer non-formal knowledge, understanding so integrated into routine performance that the practitioner stops being aware they're using it. A trainee can hear the correct diagnosis, the right medication, even a clear rationale, and still walk away without the thing that actually made the explanation work: the years of exposure that let the senior clinician arrive there instantly.

This is not a failure of teaching effort. A trainee who gets a clear, coherent explanation is receiving an accurate account of a conclusion, not a map of the reasoning that produced it. If a correct answer alone were enough to transfer judgment, decay and teaching would be two different problems. They aren't. Why this happens, and why the most experienced clinicians are often least equipped to verbalize what makes them valuable, is covered in what gets lost when senior doctors teach.

Most visibly, in intuition itself

Clinical intuition is tacit knowledge in its purest form, and the place where the explicit/tacit divide is easiest to see. Gary Klein's Recognition-Primed Decision Model, built from studying expert judgment under real time pressure, found that experienced practitioners don't weigh options analytically in the moment. They recognize a situation as belonging to a known pattern and act, with the recognition happening below the level of conscious access. Nurses across specialties have independently coined shorthand for exactly this: "JDLR," just don't look right, borrowed from policing; "ADR," ain't doin' right, from veterinary work; "STAR," shit that ain't right, from a NICU. Three different fields landing on the same phrase for the same unnamed thing is itself a kind of evidence, a term invented specifically because the underlying judgment resists more precise language. Eva and colleagues found the same thing from the other direction: expert clinicians use step-by-step analytical reasoning significantly less often than novices in diagnostic tasks, not because they've abandoned reasoning, but because experience has automated it past the point of voluntary reporting.

A pediatric cardiac ICU nurse described this mechanism almost exactly as the research frames it: intuition, she said, is usually the subconscious brain recognizing patterns and cues the conscious brain is too occupied to register. She'd been caring for a child with complex, tenuous cardiac anatomy, stable but not thriving, and on one shift something about him felt wrong in a way she couldn't immediately name, before any monitor confirmed it. He looked gray, was inconsolably agitated, and nothing on exam explained it. Minutes later his oxygen saturation dropped precipitously; a shunt had failed, and the team caught it in time. A clinician who reads a deteriorating patient before the numbers move isn't withholding an explanation. They have genuinely lost access to the intermediate steps, because those steps compressed into a single automatic response somewhere across thousands of prior cases.

What happens when that intuition gets overridden shows the cost of treating it as unreliable just because it can't be justified on paper. A home health nurse described flagging a patient as altered for three consecutive days, repeatedly telling EMS something was wrong even as vitals, labs, and imaging kept coming back normal. Twice, the call was made not to transfer her. On the third day, with the patient now barely responsive, an experienced paramedic overruled the wait-and-see approach on the nurse's word alone. The hospital workup found an ongoing heart attack with none of the standard symptoms, caught only after the damage was already extensive. What this means for how, and whether, intuition can be taught is explored in why clinical intuition is hard to teach.

Why this is one problem, not four

Go back to the resident, the specialist, the professor, and the clinician who reads a room before the monitor does. Run each of them through the same lens and the same fact repeats. The knowledge that actually matters, the part that determines whether a clinician reasons well, teaches well, or reads a patient accurately, lives in a compressed, tacit layer built through direct exposure. And in each case, the tool being used to manage it is explicit: awareness training for the resident's bias, refresher courses for the specialist's decay, verbal explanation for the professor's teaching, self-report for the clinician's intuition.

These tools don't fail by accident. They're aimed at the wrong layer. Awareness of a bias is explicit content sitting next to a tacit process it can't reach in time to intervene. A refresher course delivers explicit content to a gap that is mostly tacit erosion. An explanation transmits the account of a judgment, not the process that generated it. Self-report asks a clinician to narrate something that expertise has, by definition, made unnarratable.

What looks like four training problems is one structural fact: most of what makes a clinician expert was never stored as content, so it cannot be delivered as content. It was built, case by case, into a pattern library operating faster than conscious description can follow. Each of the four problems above has, at some point, been treated as a communication failure: a clearer bias checklist, a better-explained teaching round. The evidence says otherwise. The gap isn't in the explaining. It's in the model of what's being explained.

What actually builds this kind of knowledge

If tacit knowledge can't be delivered through instruction, the practical question changes. Not how to explain clinical judgment better, but how to design the conditions it actually forms under.

The research points to a few consistent levers. Deliberate exposure to a wide range of presentations builds the pattern library faster than passive study. Corrective feedback arriving soon after exposure lets that library update accurately instead of reinforcing an early wrong impression. Structured reflection after a case has resolved shows more consistent effects than any attempt to interrupt reasoning mid-case.

Clinicians already reach for this instinctively, in the absence of anything built for it: invite-only groups gated by license verification, specialty-wide chat threads, pathologists mailing slides across the country to the one or two people who see enough of a rare pattern to recognize it. One doctor put the routine plainly, urgent cases mean calling a colleague, routine ones go through a closed group chat, a workaround built around exactly what curricula can't deliver. Formal attempts haven't consistently closed that gap either. One credential-verified physician platform, built specifically for this kind of exchange, gets described by its own users as thinner than an open, unverified subreddit. Verifying someone is a licensed clinician solves the trust problem. It doesn't solve the harder one: getting the right case, at the right moment, in front of someone whose reasoning is worth learning from. Medcoterie, a product built inside Curolynk's own work on this question, starts from that distinction rather than assuming credential-gating alone is the fix.

None of this is primarily verbal. It's closer to curation than to teaching, less lecturer, more editor. The value isn't in narrating the footage, it's in choosing which footage a trainee sees, in what sequence, with what feedback attached. The senior clinician's most valuable role here isn't explaining more clearly. It's choosing well.

What this means for how clinicians train

None of this makes clinical judgment harder to build. It makes clear what actually builds it — which, if you're the one trying to build it, is the more useful thing to know. If you've ever struggled to explain why a case felt wrong, that wasn't a failure of articulation. If you've ever given a trainee your clearest explanation and watched it land as somehow incomplete, that wasn't a failure of teaching either. You were both running into the same wall from opposite sides.

The more useful question isn't how to explain better. It's how to make your exposure, your feedback, and your reflection more deliberate — for yourself, and for whoever you're training. That's a design problem, not a communication one, and it has a different set of answers.

Frequently asked questions

What is tacit knowledge in clinical medicine, and how is it different from explicit knowledge?

Explicit knowledge can be stated directly and transferred through teaching, like dosing guidelines or diagnostic criteria. Tacit knowledge builds through direct clinical experience and compresses into pattern recognition that a clinician can use but not fully verbalize. Most of what constitutes clinical judgment, as opposed to clinical knowledge, sits in the tacit category.

Can clinical judgment be taught directly, or does it have to be built through experience?

The evidence points to experience-dependent construction over direct instruction. Explicit teaching transfers the frameworks and vocabulary clinicians use to make sense of what they see, but the pattern recognition that produces fast, accurate judgment builds only through repeated exposure to real or realistic cases, with feedback that lets the pattern library correct itself over time.

Why doesn't teaching clinicians about cognitive bias reduce diagnostic errors?

Because biases operate in fast, automatic reasoning, while awareness of them requires slow, deliberate reasoning that's only partly available under real clinical conditions. By the time a clinician consciously considers whether they're biased, the biased judgment has usually already shaped what evidence they attended to. Structural changes to clinical environments and reflective practice after cases show more consistent effects than awareness training alone.

What actually helps trainees build clinical judgment faster?

Deliberate exposure to a wide range of presentations, timely corrective feedback, and structured reflection after a case has resolved all show consistent positive effects in the research. These are design choices about what a trainee is exposed to and how they're supported afterward, not content that can simply be explained more clearly.

References

1. Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux.
2. Polanyi, M. (1966). The Tacit Dimension. University of Chicago Press.
3. Croskerry, P. (2009). A universal model of diagnostic reasoning. Academic Medicine, 84(8), 1022–1028. https://doi.org/10.1097/ACM.0b013e3181ace703
4. Norman, G.R., Monteiro, S.D., Sherbino, J., Ilgen, J.S., Schmidt, H.G., and Mamede, S. (2017). The causes of errors in clinical reasoning: cognitive biases, knowledge deficits, and dual process thinking. Academic Medicine, 92(1), 23–30. https://doi.org/10.1097/ACM.0000000000001421
5. Ebbinghaus, H. (1885/1913). Memory: A Contribution to Experimental Psychology. Teachers College, Columbia University.
6. Custers, E.J.F.M. (2010). Long-term retention of basic science knowledge: a review study. Advances in Health Sciences Education, 15(1), 109–128. https://doi.org/10.1007/s10459-008-9101-y
7. Eraut, M. (2000). Non-formal learning and tacit knowledge in professional work. British Journal of Educational Psychology, 70(1), 113–136. https://doi.org/10.1348/000709900158001
8. Bereiter, C. and Scardamalia, M. (1993). Surpassing Ourselves: An Inquiry into the Nature and Implications of Expertise. Open Court Publishing.
9. Klein, G. (1998). Sources of Power: How People Make Decisions. MIT Press.
10. Eva, K.W., Hatala, R.M., LeBlanc, V.R., and Brooks, L.R. (2007). Teaching from the clinical reasoning literature: combined reasoning strategies help novice diagnosticians overcome misleading information. Medical Education, 41(12), 1152–1158. https://doi.org/10.1111/j.1365-2923.2007.02923.x