Skilled diagnosis
A Socratic walk-through of skilled diagnosis — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #How does an experienced technician notice the source of a fault before a novice does?
The natural assumption is that the experienced hand knows more — a longer catalogue of faults and their fixes. But watch one work and the assumption creaks. The expert often has the answer before finishing the information-gathering the novice thinks is required, sometimes before touching anything. If expertise were a bigger lookup table, it would still need the same inputs to look up. So the advantage sits upstream of the reasoning. Where?
Reasoning it through
REASONING #It sits in perception, and a demonstration from chess isolates the effect cleanly. Show a master a real position for a few seconds and they reconstruct it far better than a weak player; scatter the same pieces at random and the advantage largely evaporates. Chase and Simon drew the conclusion: the master has no better memory, but sees configurations where the beginner sees pieces — and a configuration is one thing to hold rather than twenty. At a machine, the novice hears a noise while the expert hears a bearing: same air pressure at both ears, but one has a category ready, and it arrives with its implications attached.
Where do those categories come from? Here is the surprising part. To hear that a machine sounds wrong, you must already carry an expectation of how it sounds when right. The expert's perception is comparative — not "loud", but "loud in that one harmonic, only under load" — so most of what built the expertise was not faults at all, but hundreds of hours around healthy machines, laying down a baseline nobody was consciously studying.
That baseline buys what a symptom list cannot: the ability to notice an absence. The pump that should cut in and does not; the click that should follow the relay and never comes. Consider how invisible an absence is to someone with no expectation — there is nothing to attend to. Much of skilled diagnosis is noticing what failed to happen.
A second selection runs alongside it. The novice attends to what is loudest, salience being all they have; the expert attends to whatever separates the candidates — often a dull observation not on the novice's list. Klein, studying fireground commanders under time pressure, found they rarely compare options: recognition hands them one plausible answer, which they run forward mentally to see whether it holds. And that running-forward has something to run on — a working model of how the machine causes its own behaviour, turning three symptoms into one constraint. The question stops being "what causes a warning light?" and becomes "what single fault could produce all three, in this order?"
The analogy
THE ANALOGY #Think of a fluent reader. They do not assemble words from letters; the word arrives whole, and a misspelling jumps off the page before the sentence has consciously been read. The beginner, sounding out letters, must reach the end of the word to know what it was — and cannot notice a typo at all, having no expectation to violate.
written language is a closed, fixed code where fluency converges on the truth, whereas machines are open — a familiar pattern can have an unfamiliar cause — so the fluency that makes the expert fast will occasionally read a novel fault as the word it usually is.
Clarifying the model
THE MODEL #That last sentence is not a decorative caveat but the occupational hazard. The recognition that makes an expert fast can close the question too early, and the countermeasure good practitioners adopt — a habit, not a talent — is keeping a second hypothesis alive and asking what evidence would kill the first.
Two further honesty points. Skilled intuition is only trustworthy in domains regular enough to contain real patterns and giving feedback fast enough to learn them — Kahneman and Klein converged on those conditions. Where outcomes are delayed, noisy, or never checked, confident pattern-recognition can be no better than a coin toss while feeling identical from inside. And none of it transfers: the apparatus is built from one machine's normal behaviour, so an expert on strange equipment is a novice with good habits.
A picture of it
THE PICTURE #How to readRead the sections left to right as stages of one callout, and each line's score as how much traction that person has at that moment rather than how happy they are. The lines within a section are the same instant experienced twice, so the gap between them is the whole subject — and it opens widest in "first impressions", before any reasoning, which is the evidence that the advantage is perceptual rather than deductive. The final one-line section is the honest ending: move the expert to equipment whose normal behaviour they never absorbed and the score drops, the baseline having done the work.
What became clearer
WHAT CLEARED #The expert's advantage arrives before the thinking starts. Long exposure to normal machines builds an expectation precise enough that a deviation — including a missing event — presents itself as a named thing rather than raw sensation, and recognising it supplies a candidate explanation to be checked rather than trusted. Not a bigger list of faults, but a better-calibrated sense of what should have been.
Where to go next
ONWARD #- Why teaching the normal state, not the fault catalogue, may be the faster route to competence.
- What becomes of this expertise when the diagnostic signal moves from sound and vibration to a screen of data.
Key terms
TERMS #| Term | What it means |
|---|---|
| Chunking | the perceptual grouping of many elements into one meaningful unit, letting an expert take in a scene at a glance. |
| Recognition-primed decision | Klein's account in which experts recognise a situation type, generate one course of action, and simulate it rather than compare alternatives. |
| Premature closure | settling on the first plausible explanation and ceasing to look; recognition's failure mode. |
Every term the collection defines is gathered in the glossary.