Critical Incident Technique¶

Method — instantiates Tacit Knowledge Elicitation

Collects and dissects the rare high-information episodes — successes, failures, and near misses — where routine behavior broke, because those are the moments that reveal the judgment ordinary descriptions leave out.

On a normal day, expertise is invisible — everything works, and the expert can tell you nothing useful about why. Critical Incident Technique deliberately hunts the abnormal day: it gathers the specific episodes where something went unusually right, unusually wrong, or nearly wrong, and mines them for the judgment that only becomes visible under strain. Its defining move is selecting on information, not on representativeness — it goes looking for the near miss and the save precisely because those are the moments when the tacit rule was tested, bent, or broken, and therefore surfaced. The unit is the incident: what made it critical, what the practitioner noticed, what they did, and — the heart of it — how their behavior departed from the routine that a manual would prescribe.

Example¶

A rail traffic control center runs a critical-incident study to capture what its best dispatchers know. Instead of asking "how do you route trains?", investigators collect specific near-misses and saves: the shift where a dispatcher held a freight train on a hunch and a track defect was found an hour later; the near-collision that was caught in the final seconds; the time a standard re-routing would have been correct by the book but a veteran overrode it.

Each incident is dissected for the exception: what was the first sign the ordinary procedure didn't fit? In the held-freight case, the dispatcher recalls a pattern of small sensor anomalies that individually meant nothing but together "smelled wrong" — a discrimination no procedure encodes. Across a dozen such incidents, a pattern of tacit judgment emerges: the conditions under which veterans deviate from the standard playbook and why those deviations were right. What the study deliberately doesn't do is describe the ninety-nine ordinary shifts; its whole leverage comes from the rare ones, which is also its main hazard.

How it works¶

Select on criticality. Gather episodes chosen because they are high-information — clear successes, real failures, near misses, hard edge cases — not a representative sample of ordinary work.
Reconstruct the incident concretely. For each, recover what happened, what the practitioner attended to, what they did, and what the outcome was, in enough specificity to see the judgment.
Probe the departure. The core question is where and why behavior diverged from the routine or the rule — the exception is where the tacit expertise lives.
Look across incidents for pattern. Recurring departures across many incidents reveal the conditions under which experts safely break the standard — the reusable finding.

Tuning parameters¶

Incident sourcing — self-reported memorable cases, or incidents pulled from records and logs. Self-report is fast but memory-biased; record-sourced incidents are more objective but costlier to reconstruct.
Success/failure balance — how many saves versus failures versus near-misses. Failures are vivid and well-documented; near-misses are the richest and most under-reported; a lopsided mix skews the lesson.
Criticality threshold — how dramatic an episode must be to qualify. A high bar concentrates on rare extremes; a lower bar admits everyday-hard cases that generalize better.
Attribution stance — blame-oriented or learning-oriented framing. A learning frame recovers far more honest exception data, especially for failures and near-misses.

When it helps, and when it misleads¶

Its strength is leverage: a handful of critical incidents can reveal tacit judgment that a hundred routine observations would never expose, because the exception is where hidden rules become visible.^[1] It is uniquely good at surfacing when experts correctly break the rules.

Its central failure mode is overlearning from the dramatic. Selecting on rare, memorable incidents skews the sample toward the extreme; a team that studies only near-disasters can end up over-fitting to spectacular cases and mis-weighting the everyday judgment that keeps normal days normal. Self-reported incidents add hindsight and memory bias, and a blame-tinged framing quietly suppresses the honest exception data — especially around failures — that the method most needs. It is also easy to misread a survivor's account: the save that "worked" may have been luck, not skill. The discipline that keeps it honest is to balance the incident set against a base of ordinary cases (so the exception is read against the norm, not instead of it), source incidents from records where possible, and hold a strictly learning-oriented, non-punitive frame so near-misses actually get reported.

How it implements the components¶

Critical Incident Technique fills the components concerned with rare, rule-breaking episodes:

critical_incident_case — its unit of work: the deliberately selected high-information success, failure, or near-miss, reconstructed for what it reveals.
exception_probe — its core question: where and why practice departed from the routine or the rule, which is where tacit judgment concentrates.

It surfaces incidents but does not store or curate them as a corpus (provisional_articulation, transfer_boundary_annotation — Case Library Capture), does not observe live practice (expert_practice_observation — Shadowing Session), and does not build the whole-task model (decision_rationale_probe across the task — Cognitive Task Analysis).

Instantiates: Tacit Knowledge Elicitation — Critical Incident Technique supplies the high-information exception cases the rest of the elicitation reasons from.
Sibling mechanisms: Case Library Capture · Decision Trace Review · Cognitive Interview · Cognitive Task Analysis · Expert Debrief · Think-Aloud Protocol · Shadowing Session · Apprenticeship Observation · Simulation Replay · Judgment Aid · Task Analysis

Notes¶

Critical Incident Technique typically feeds Case Library Capture: the incidents it surfaces are exactly the high-value entries a case library wants to store. Keeping the two distinct matters — the technique's job is to find and dissect the exception; the library's job is to hold it, alongside enough ordinary cases to keep the collection from becoming an anthology of disasters.

References¶

[1] Building an account of effective and ineffective performance from collected accounts of concrete, decisive incidents is the Critical Incident Technique introduced by John Flanagan (1954), which grew out of wartime studies of aircrew performance. Its selection-on-criticality is the source of both its leverage and the sampling-bias caution above. ↩