Everything in this document is a derivation of that sentence. Whole diagnostic reasoning compresses into four verbs: organise, discriminate, decide, learn. When you can perform those four on any presentation, you have the skill. When you name the wrong verb, you have diagnosed the exact fault that cost you the question.

1. Organise — the complaint into a tree of families (where the process lives).
2. Discriminate — apply the feature that separates siblings at the fork you are on.
3. Decide — pick the leaf that survives, then test it against the full data.
4. Learn — log the fork and the discriminator, not the answer, so the tree grows.

“It’s very orderly. Family, sub-family diagnosis. Shortness of breath starts with heart and lung, so now I’ve got my chief complaint. That’s my organ system, and now we do our sub-family.”

— the reasoning you asked me to codify

A diagnosis is a hypothesis, not a truth

You never prove a diagnosis the way mathematics proves a theorem. You never hold all diagnostic possibilities. You hold a provisional best-supported explanation, because medicine is committed to one action at a time rather than a completeness requirement. This is a quiet, profound release: you are not trying to be right; you are trying to be least-reasonably-wrong, then to act.

The combinatorial trap

There are on the order of thousands of diagnostic entities a clinician could name. If diagnosis were enumeration — compare each presentation to every disease — it would be impossible and the differential would never close. That is precisely why the shotgun approach fails: memory is finite, and pattern-match works only for the already-known leaf, not for an unfamiliar stem.

Why the hierarchy is the escape

The hierarchy defeats enumeration because it converts one comparison against many into a few comparisons, done one after another.

Medicine’s disease space is naturally and deeply hierarchical — organ → mechanism → pathology → entity — because anatomy is real, physiology is real, and disease is real. You need not invent the tree; it is encoded in the body and in every textbook’s chapter structure. The work of this framework is to make you use that hierarchy at speed instead of hauling the whole list to every question.

The recursion and its five levels

One constant and a few lines of rule: the tree is recursive — every node is examined exactly like its parent (identify children, discriminate among them, descend). The levels we always begin with:

1. L0 — The complaint. The chief complaint in the patient’s own words. No diagnosis here. Just a human who says “short of breath.”
2. L1 — The organ / system. Where the process lives. Anatomy. SOB → heart & lung (never wide open).
3. L2 — The mechanism family. How the organ fails to convey the symptom. Heart → pump failure, timing, obstruction, or filling?
4. L3 — The sub-family. A smaller call within a family. Pump failure → left versus right ventricle.
5. L4 — The disease entity. The diagnosis, the leaf you commit to. Aortic regurgitation, not just “left valvular disease.”

Are five levels essential? No. The recursion is essential; the depth is whatever the data supports. Some stems stop at L2; most resolve by L3; a few demand the L4 entity.

The two error-binding laws: exhaustiveness and exclusivity

Every time you descend, two laws must hold at the level you are on. They are what separates discipline from a miscarriage of reasoning.

• The family must be complete (exhaustiveness). All plausible children of the mechanism are visible. If you silently drop a child — say, the diastolic causes of SOB — every question that tests it is over before you begin. Completeness is a memory requirement: the one place this framework cannot cover for you.
• The children must be disjoint (exclusivity). Each branch is meaningfully different, so one discriminator can pick one. If two children overlap you manufacture false ambiguity — no knife can settle them.

Knowledge vs. reasoning at each level

Be honest about which is doing the weight — it tells you where your real studying lives:

Level discipline — the jump that kills

The single most dangerous maneuver in the framework is the leap: dropping two or more levels at once because a pattern “felt” like something. The quotation is the warning — “still haven’t got to the diagnosis” — written to show that the leap has not arrived, and that false surety misleads.

What a discriminator is

A discriminator is a single feature (or a tiny feature-set) that separates the children of the node you are currently on. It is not a general fact; it is a local knife, sharp for exactly one fork. “Aortic vs mitral” is decided by auscultation and the waveform; “pump-fail vs obstruction” is decided by eye and pulse. Different forks, different knives.

The skill to cultivate is matching: which discriminator applies at this particular node, from the several you know. Almost every stem is an invitation to use exactly one discriminator — find the writer of the invitation.

Where discriminators live — the evidence stack

A discriminator lives in history, a physical sign, or a test. Prioritise by reliability and availability:

1. History (what the patient says; the stem’s narrative) — always present, always the anchors.
2. Physical-exam signs — the classic wide-pulse-pressure sign, the murmur, the clubbing collar.
3. Investigations — the result that ends a fork (echo, ECG, chest film) — only if handed to you.

Discriminator priority

Not all discriminators are equal, and the exam almost always honours one that is high-priority. Favour features that are

• Specific (near-pathognomonic beats generic),
• cheap and present (present in the vignette — the writer had to put it there),
• directly at the fork.

The one-fork principle

Also deliberate, and central to the excerpt: the walk passes through many forks, and a stem is built to spare you all but one. The question’s real discriminator sits at exactly the fork that separates the two most similar answer leaves.

Disguises and reverse discriminators

The exam’s favourite move is to disguise an answer so it reads as the wrong leaf — through a synonym, a look-alike presentation, or a distractor from another branch. These are the accurate answers that are still the wrong leaf.

• Wrong-family disguise. Plant an L1/L2 answer inside a vignette that really tests L4.
• Disguise by synonym. Use the everyday word instead of the exam’s phrase (describe the finding rather than name it).
• Ambiguity for hire. Load evidence pointing at two forks at once, so the discriminator seems blurred.

The counter-move is always the same: name the fork. If you can state “the fork is valve vs pump” and “the discriminator is the murmur location and pulse pressure,” the disguise is dismantled.

Read the stem as a fork being rebuilt

Do not read a stem as prose; read it as data for the tree. As you read, assign each sentence to a level:

1. Extract the complaint (L0) — the anchor, fenced without eager translation.
2. Drift the candidate organs (L1), one pass.
3. Let the narrative features settle a mechanism (L2).
4. Settle sub-family and entity (L3/L4) with the discriminating sign.
5. Class each option against the fork, not against memory.

The four ordered operations

Apply them in order; never reorder. Together they map the tree onto the offered set of answers.

1. Position. For each option, identify which level it represents. This is the act most students skip — and the single biggest return on invested time.
2. Open the fork. Rebuild, once, the fork the discriminator is sitting on, cleanly.
3. Select the survivor. Apply the discriminator to the children and see which option is still standing.
4. Test the whole stem. Confirm the survivor reconciles every sentence — the whole-stem check catches an option true of several sentences yet wrong on the decisive one.

Cognitive task — which lane the answer lives in

Not every question uses the whole tree. Label the cognitive demand before you attack:

Knowing the lane is itself a decision fork. Treat an analysis question as recall and you will reach for a fact and jump a level. Treat recall as analysis and you will dissolve a single fact into a debate. Label the task, then score.

The trap taxonomy

Why every distractor is wrong

When you debrief, produce a reason for each distractor, in a single phrase:

1. State the distractor’s level and its fork.
2. Say the discriminator that kills it (“the leak gives the wide pulse — so aortic beats mitral”).
3. Name the trap type from the table above.
4. Write a one-line future alert — the exact trigger to avoid next time.

The four error species

Every error is one of four species. Classify honestly — the treatment differs radically:

1. Level error. You answered the fork above or below the test's. Fix: level discipline. The easiest to cure once named.
2. Discriminator error. You had the fork but misapplied or misread the separating feature. Fix: restudy the sign’s identity and location.
3. Attraction error. You chose the seductive jump — you felt it and went. Fix: habit; slow the leap.
4. Fact error. The branch was never in your tree — exhaustion failed. Fix: add the branch; often a whole family is missing.

The tree audit

Periodically, look back at a batch of errors and ask: which branch of which system’s tree is simply absent? If all your misses cluster at the same missing child of, say, the jaundice tree, then restudying “jaundice facts” is the wrong fix — you must add the missing child to the tree. Audit by tree node, not by subject name.

The error-log record

Log the full record for every wrong answer — this is the file you will mine for the audit:

Build the tree before the facts

For each high-yield system, open a per-system template (Section 9) and draw the tree before you read a single fact. Start from the chief complaints that open that system. The scaffolding then tells you where the facts go — you are building the frame to hang the leaves on, the opposite of reading the chapter cold.

Flashcards belong at the fork, not the leaf

The single best study optimisation in this framework: make the cards at the fork.

• Card — fork ask: “Pump vs obstruction — which discriminator settles it?” not “what is heart failure?”
• Card — sign → fork: “wide pulse pressure → which fork?”
• Card — mechanism → signs: “AR mechanism → what exam signs?”
• Card — two leaves apart: compare AR and AS by every discriminator.

Leaf-only cards teach recall, not discrimination — and discrimination is what is tested under time.

The three daily loops

1. Build / refresh. Maintain the tree for the day’s system.
2. Practise. A small intentional batch of real MCQs, run through the MCQ machine.
3. Harvest. Submit every miss to the error engine: order the record, classify the species, log it.

Tie to weighting and cycles

The framework is the machinery; the syllabus weighting and your exam cycle are the itinerary. Fit the systems’ trees to the rotation / cycle map — a full cycle’s 200 MCQs maps onto the per-system trees. Sum the cycle’s results by system, by tree node, and by error species — those three per-axis summaries are exactly what the framework is built to hand back.

The whole machine on one stem, every step written out — the dyspnoea walk from the framework’s motivation, with the answer-tactics now applied.

1 · Extract the complaint (L0). “Breathlessness on exertion.” Preserve the words; do not yet interpret.

2 · Open L1 — organs. SOB → heart first, lung second. The stem carries cardiac clues (pulses, murmur), so lead with heart, station lung.

3 · Mechanism (L2). The wrist findings (wide pulse pressure, Corrigan’s) point to a valve-competence / ejection dynamic, not simple pump failure. The murmur and pulse pull away from pure-pump failure toward a structural valvular mechanism.

4 · Sub-family and entity (L3/L4). The discriminating sign is the pressure plus the auscultatory finding. The leaf is aortic = aortic regurgitation, not aortic stenosis (which would give a narrow pulse / systolic murmur). The fork here is AR vs AS; the discriminator: wide pulse + early-diastolic murmur vs narrow pulse + systolic crescendo–decrescendo murmur.

5 · Class the answers by level and kill. Given: A) aortic stenosis, B) aortic regurgitation ✓, C) mitral regurgitation, D) left-ventricular failure —

Duplicate this outline once per high-yield system. Fill it first from the syllabus and past-paper patterns; hang the facts after. Copy this out by hand — the template is a scaffold for your own systems: SOB, chest pain, jaundice, fever, abdominal pain.

The whole framework, compressed to twenty reusable rules.

1. Diagnosis is never enumeration; it is tree descent.
2. Name your level aloud (L0–L4) before you choose any leaf.
3. Never skip a fork — the leap is the exam’s favourite.
4. Begin each differential at the organ; anatomy is the first split.
5. Complete the family (exhaustiveness) — the dropped child is the commonest fact error.
6. Keep the children disjoint (exclusivity) so one knife can settle them.
7. Every point has one active fork from its minimal pair.
8. The discriminator is local to the fork; it is not a general fact.
9. Treat included data as the writer’s confession of where the discriminator sits.
10. Classify the option by level first (position), then rule it against the fork.
11. An answer true but for a different branch is the classic distractor.
12. “Heart-failure”-type leaves (the L2 you feel) are the seductive jump.
13. Classify the cognitive task (recall / interpretation / analysis / synthesis) before scoring.
14. Even a correct answer earns a distractor-hunt: why is every option wrong?
15. Log the full error record: pair, level, species, future alert.
16. Attribute each error to one species: level / discriminator / attraction / fact.
17. The audit question — which tree node is absent? — decides whether you open a chapter or do MCQs.
18. Build the tree for a system before you read the facts.
19. Flashcards at the fork: drill discriminators and sign-to-fork ties, not leaf definitions.
20. Practice + harvest must be near-daily. The tree is a skill that compounds only through repetition.