The Decompression Order

Tuesday evening sign-out. A senior resident is handing off a patient to the overnight team. Same patient, same facts, two possible openings.

Opening A: "Mrs. K, 47, uncomplicated appendicitis on CT, surgery aware. NPO, on Zosyn, scheduled for OR at 23:00. Pain controlled with PCA. Call if she develops new rigidity or fever above 38.5; otherwise expect the surgical team to take her at 11 PM."

Opening B: "Mrs. K is a 47-year-old who came in this afternoon with right lower quadrant pain. Labs show WBC 14, lactate 1.2, K and creatinine normal. CT shows appendicitis with no perforation. She has been NPO since arrival, on Zosyn, pain controlled with PCA. Surgery has been consulted and accepts. They have scheduled her for OR at 23:00."

Same content. Opening A loads the diagnosis first and the receiver's appendicitis model activates immediately; the NPO status, the antibiotic, and the OR time then modulate that model in the order the receiver needs them; the precautions land last, where they will sit on top of an already-built picture. Opening B forces the receiver to read every line waiting for the diagnosis to emerge. By the time "CT shows appendicitis" arrives, the receiver has been holding a stack of unbound facts in working memory. They now have to retroactively bind each fact to the model. The retroactive binding is doable, but it is more cognitive work, and the work happens at sign-out, which is the worst possible moment to spend cognitive budget.

The principle

Once the abstraction level is chosen (Part 3 did this), the message should unfold in a fixed order. Base model first. Relevant nuance second. Evidence and action third.

The three questions the order answers, in this order:

What is the patient? What matters about this version of the model? What needs to happen?

This is the order the receiver's cognition runs anyway, whether the message helps or not. The message that fits the cognition saves the receiver work. The message that fights the cognition costs them work.

Why this order

The base model lets the receiver orient. Once "DKA" lands, the receiver knows the framework: dehydration, acidosis, insulin, potassium watch, gap closure, transition criteria. Until then, they are holding raw facts with nowhere to put them. The model is the container; the rest of the message goes inside it.

The relevant nuance prevents overgeneralization. "DKA" alone activates the whole framework, but this DKA is improving, this potassium is borderline low, this presentation is severe enough to need an ICU bed. The receiver's stored model contains the average DKA, not this DKA. Nuance specifies which version of the framework applies. Without it, the receiver applies an average that may be wrong.

The evidence and action tell the receiver what changes management. After the model is loaded and modulated, the receiver needs to know what they are doing next. "Continue insulin drip with potassium repletion before transition" lands cleanly because the model and nuance are already in place; the action sits on top of a built picture. Out of order, the same action is ambiguous: continue what insulin drip, transition to what, repletion of what potassium.

The order is receiver-relative. A trained clinician decompresses fastest with model-first; a database does not care about order at all; a layperson may need the action first ("she needs to stay overnight") before the model and nuance make sense. The unfolding pattern in this essay is the one that fits the cognition of a clinician who shares the codec. When the receiver is a different substrate, the order has to be re-derived from what that substrate decompresses cheaply.

What goes wrong with the opposite order

Two common inversions of the order.

Evidence-first. The clinician (or the AI) leads with labs, vitals, history, exam findings, and only arrives at the diagnosis or syndrome at the end. The receiver sits in working memory holding unbound facts until the model lands. By the time the model arrives, the early facts are partly forgotten and have to be back-bound. The chart note that runs "HPI, ROS, exam, labs, imaging, assessment, plan" inverts the order; the assessment that the receiver actually needs sits below five sections of evidence they had to read first.

Action-first. The clinician leads with the action ("started on insulin and fluids") and lets the receiver infer the model from the action. This works between clinicians who share a strong codec (the listener hears "insulin and fluids" and infers DKA), but it fails for any receiver who does not share that codec: the cross-covering clinician, the consultant from a different specialty, the family member, the AI. Action-first compression assumes the receiver has the same model the sender does; when they do not, the action looks unjustified.

Both inversions share a structural feature. They force the receiver to do the reconstruction work the sender should have done. The unfolding order is part of the compression; getting it wrong means the message has more facts and less function.

Variation by function

Part 2 named several functions clinical communication performs. The order shifts slightly across functions, but the underlying pattern holds.

For handoff, the base model is the diagnosis or syndrome, the nuance is trajectory and outstanding work, the action is what to call about. For a consult request, the base model is the patient summary, the nuance is the decision context, and the action is the specific question being asked of the consultant. For a chart note, the base model is the assessment, the nuance is the supporting evidence, and the action is the plan; the conventional SOAP order (subjective, objective, assessment, plan) inverts this and is one of the most studied examples of order-against-cognition in clinical writing. For teaching, the base model is the reasoning rule, the nuance is the discriminating features, and the "action" is the generalization to the next patient.

The principle survives the variation. Whatever the function, the message that loads the model first, modulates it second, and acts on it third is the one the receiver can absorb at the rate they need to absorb it.

The clinical AI implication

A clinical AI output inherits the same order. The model the system has built should come first. The nuance for this patient and this moment should modulate it. The action implication should land last.

A copilot that returns a wall of evidence and lets the clinician infer the model is making the clinician do work the system could have done. A scribe that produces a chart note in the conventional SOAP order is inheriting an inversion the field has accepted but does not need to. A differential generator that returns a ranked list of diagnoses with no nuance and no action implications is producing output that the clinician must finish before it becomes useful.

The system that respects the decompression order produces a different shape of output. The leading sentence names the model. The next two sentences specify what matters about this version. The closing names what changes management. The clinician reads top to bottom and the message lands fully on first read.

At the bedside

The senior resident in sign-out used Opening A. The overnight team took the patient, called the OR at the right time, and watched for the precautions that mattered. Nobody had to rebuild the picture from the evidence. The message did the binding work, in the order the receiver's cognition was going to do it anyway.

The next two parts handle what to leave out (Part 5) and what to keep when in doubt (Part 6). Both depend on this one: the right order first, then the right selection within that order.

---

Compressed Medicine · Preface · 1. The Compression Substrate · 2. The Function of the Message · 3. The Highest Accurate Abstraction · 4. The Decompression Order · 5. The Minimum Sufficient Message · 6. The Grounding Constraint · 7. The Belief-State Object · 8. The Same Wall · 9. The Defense Architecture · 10. The Temporal Loop · 11. The Irreversible-Action Check