Every module in this course, up to now, has been describable as a discrete step you could deliberately run: build the differential, audit the bias, rate the confidence, check the applicability. That's what makes them teachable. But the clinicians most worth learning from don't visibly run twelve separate steps on a routine case — the steps have compressed into something that looks, from the outside, like intuition. Modules 51 through 55 are about that compression: what it actually is, how to capture it before it's lost, and how to tell it apart from mere shortcut-taking, which looks identical from the outside and is dangerous in exactly the way real expertise isn't.
This brings the course back to where Lesson 1 started — expertise compresses itself, and the compression is what makes it hard to teach. Twelve lessons later, the goal isn't to reverse that compression permanently. It's to have a fallback: the ability to decompress your own reasoning back into its explicit steps whenever a case is unusual enough, or the stakes are high enough, that intuition alone isn't good enough to trust.
The goal was never to make you run twelve explicit steps on every routine case forever. It was to build the habit deeply enough that you can drop back into it deliberately on the cases that actually need it — and to know which cases those are.
This is the difference between a cached result and a genuinely optimized model — both return the answer fast, but only one of them can still show its work when the input falls outside the cases it's seen before.
Module 51 — Tacit Knowledge Elicitation
Module 51 exists because expert clinicians routinely cannot answer "how did you know" in a way that transfers to a learner — not from unwillingness, but because the knowledge itself has become procedural rather than declarative. The module's technique is indirect: instead of asking the expert to explain their reasoning directly, it asks them to react to a series of deliberately varied hypothetical cases, and infers the underlying rule from the pattern of their reactions rather than from their stated explanation, which is often an after-the-fact rationalization rather than the actual mechanism.
If you've ever tried to teach a skill and found your own explanation didn't match what you actually did on the next case, you've already discovered why this module asks indirectly rather than directly.
This is probing a black-box model's decision boundary through controlled perturbation rather than trusting its self-explanation — the same reason interpretability research rarely relies on a model's own stated reasoning alone.
Module 52 & 53 — Expertise Versus Shortcut
Compressed intuition and a shortcut that skips necessary steps can look identical from the outside — both produce a fast, confident answer without visible working. Module 52 provides the diagnostic test: expertise, when decompressed under Module 51's technique, reveals a coherent rule that generalizes correctly to the hypothetical variations; a shortcut, decompressed the same way, reveals a rule that's actually narrower than the expert believes, and fails on at least one of the varied cases. Module 53 then asks the harder practical question — given a specific fast judgment, is there time and stakes enough to actually run the decompression check before acting on it, or does the situation demand trusting the fast judgment as-is?
The clinicians who get hurt by their own intuition aren't usually the inexperienced ones — they're experienced clinicians whose intuition has quietly drifted into a shortcut on one specific case type, and who never check because it's worked every time so far.
This is distinguishing a model that has genuinely learned the underlying function from one that has memorized the training distribution — both score well in-sample, and only out-of-distribution testing (Module 52) tells them apart.
Worked Example: Decompressing a Fast Judgment
An experienced clinician glances at a patient's chart and says "this one's not sick" within seconds, before formally examining anything — a version of pattern recognition from Lesson 1's opening loop, now running at a speed that skips visible steps entirely.
Asked to react to five hypothetical variations of the same presentation, the clinician's answers change consistently with respiratory rate and skin color, and barely at all with the specific complaint volunteered — revealing the actual underlying rule: it's a rapid physiologic read, not a diagnosis-driven one.
The inferred rule generalizes correctly across all five variations, including one where the presenting complaint sounded alarming but the physiologic signs didn't change — the clinician still correctly called it "not sick." That's the signature of real expertise, not a shortcut: the rule holds even on the case designed to trip it up.
Because this was a routine triage judgment and the patient was already being monitored, no formal decompression was needed before acting. Module 53's gate would have required it if the same fast judgment were being used to decide whether to discharge a patient home unmonitored instead.
Module 55 — The Closing Loop
Module 55 is the course's actual final move, and it's deliberately simple: periodically run Module 51's elicitation technique on yourself, on a case type you handle often, and check whether the rule that emerges still matches what you'd endorse on reflection. Expertise that was sound five years ago can drift into a shortcut without any single moment marking the transition — the same way a codebase drifts from its original design one small patch at a time. Module 55 is the recurring audit, not a one-time diagnosis.
That last question — would you endorse this if it weren't yours — is the whole course compressed into one line: the discipline of examining your own reasoning trace with the same rigor you'd apply to anyone else's, which is where Lesson 1 started and where the course ends.
The best clinicians aren't the ones who never need to check their own intuition — they're the ones who keep checking it, on a schedule, long after they've stopped consciously deliberating on routine cases.
This is scheduled model re-evaluation against a held-out check, not a one-time validation — the same reason a model that passed its eval a year ago still needs periodic re-testing as the world it operates in drifts.
Final Exercise: Auditing Your Own Expertise
- Pick one judgment you make often and quickly — a triage call, a "this looks fine" read, a decision you rarely explain even to yourself. Run Module 51's indirect elicitation on your own reasoning using five hypothetical variations.
- Apply Module 52's coherence check: does the rule that emerges generalize correctly to the variation designed to trip it up, or does it reveal a narrower shortcut than you believed you were using?
- Ask Module 55's closing question honestly: would you endorse this rule if a colleague described it to you cold? If not, that's the one thing from this entire course worth carrying forward and deliberately decompressing the next time this judgment comes up.
This lesson draws directly on Module 51 — Tacit Knowledge Elicitation, Module 52 — Expertise-vs-Shortcut Coherence Check, Module 53 — Stakes-Gated Decompression, and Module 55 — Periodic Self-Audit, all from the VibeRounds Prompt Directory. This closing lesson doesn't introduce a new technique to run on a case — it's a discipline for maintaining every technique in the previous twelve lessons once they've stopped feeling like technique at all. The companion Evidence-Based Medicine for Techies course pairs well with this whole course, start to finish. Neither course is a clinical decision tool; see the VibeRounds disclosure statement for full terms.