Incident Pattern Review¶
Evidence-mining method — instantiates Failure Mode Anticipation
A method that mines past incidents, near misses, tickets, and defects for recurring failure patterns, turning real base rates into likelihood estimates and observed precursors into detection signals for a new design.
Incident Pattern Review anticipates failure by looking backward at what has already gone wrong. It treats the organization's own record — incidents, near misses, support tickets, warranty returns, defect logs, complaints, audit findings — as evidence about how the next design will fail, and mines it for recurring patterns. Its distinguishing contribution among its siblings is that its outputs are measured, not imagined: it converts the observed frequency of a recurring failure into a grounded likelihood estimate (a base rate), and it extracts the detection signals — the precursors and leading indicators that actually preceded past failures — so they can be instrumented in the new design. Where a scenario review reasons forward through a hypothetical story and an FMEA scores modes on judgment, incident pattern review reads the data. Its premise is that the failures a system has already produced are the cheapest and most reliable forecast of the failures it is about to produce again.
Example¶
Before shipping a new model smart thermostat, a consumer-electronics team runs an incident pattern review on two years of the current model's history: support tickets, RMA reports, and app-store reviews. Clustering the record surfaces two dominant recurring patterns — Wi-Fi connectivity dropping after firmware updates, and false "no heat" alerts during cold-weather cold-starts — together accounting for most of the support volume. From the recurrence rate the team derives a base-rate likelihood for the analogous subsystems in the new model: connectivity failures are not a remote possibility to be scored on a hunch, they are a near-certainty at a known rate unless the design changes. Digging into the telemetry that preceded the drop complaints, they find a measurable precursor — a spike in reconnect attempts hours before users noticed — and specify it as a detection signal to instrument from day one in the new model. The review's product is a grounded frequency and a leading indicator, both drawn from the record rather than the imagination.
How it works¶
- Assemble the corpus. Pull the widest defensible record — not just logged incidents but near misses, tickets, returns, and complaints, because near misses are far more numerous than accidents and carry the same signal.
- Cluster into recurring patterns. Group the raw events into repeating classes, so the review reasons about rates rather than anecdotes.
- Compute the base rate. Turn each pattern's recurrence into an observed frequency — the likelihood input the imaginative methods can only guess at.
- Extract the precursors. Identify what measurably preceded past occurrences and hand those forward as detection signals to instrument in the new design.
- Map onto the new design. Attach each pattern to the analogous part of the new system, checking that the analogy actually holds.
Tuning parameters¶
- Corpus breadth — logged incidents only versus the full record of near misses, tickets, and complaints. Breadth multiplies the sample and surfaces silent patterns, but adds noise to sift.
- Lookback window — how far back to reach. Long windows give stable rates but drag in obsolete designs; short windows track the current system but sample thinly.
- Clustering granularity — broad categories versus fine-grained patterns. Fine clusters give specific signals but split the sample until rates become unreliable.
- Analog strictness — how similar an old component must be before its history transfers. Loose mapping reuses more data but risks importing a rate that does not apply.
- Precursor threshold — how strong a leading indicator must be to count as a detection signal, trading false alarms against missed early warning.
When it helps, and when it misleads¶
Its strength is that it is the only sibling whose frequency is measured. It surfaces the failures that actually recur rather than the ones that are fun to imagine, it corrects intuition with base rates, and it mines near misses — cheap, plentiful, and predictive — for warning signs that would otherwise be discovered only after the next incident.[n1]
Its signature failure mode is that it can only see failures that already happened and got recorded: a genuinely novel failure mode is invisible to it, and reporting culture quietly distorts the base rate, since unreported near misses simply vanish from the record and make a fragile system look safe. The classic misuse is reading an absence of past incidents as evidence of future safety for a design that is actually new. The guarding discipline is to pair the review with a prospective method for the never-seen modes, to correct openly for under-reporting rather than trusting the raw count, and to verify that the historical analog is truly analogous before transferring its rate.
How it implements the components¶
likelihood_estimate— derives each recurring failure's plausibility from its observed frequency in the historical record — a base rate — rather than from a subjective rating.detection_signal— extracts the precursors and leading indicators that measurably preceded past occurrences, ready to instrument as early warning in the new design.
It does not narrate how a single failure unfolds step by step (failure_cause, failure_effect) — that is Failure Scenario Review, its nearest analytic twin, which reasons forward through an imagined chain while this method reads real frequencies — and it does not turn its rates into a ranked score (risk_priority), which is Failure Modes and Effects Analysis.
Related¶
- Instantiates: Failure Mode Anticipation — the review supplies the empirically grounded base rates and warning signals the other mechanisms score, own, and monitor.
- Sibling mechanisms: Design Review · Failure Modes and Effects Analysis · Failure Scenario Review · Safety Case · Fault Tree Analysis · Hazard Analysis · Premortem Workshop · Risk Register
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Incident Pattern Review operates as a computation, comparison, model, or analytic representation used to infer, estimate, or choose because it a method that mines past incidents, near misses, tickets, and defects for recurring failure patterns, turning real base rates into likelihood estimates and observed precursors into detection signals for a new design
Independent corroboration: The frozen evidence defines Incident Pattern Review as 'A method that mines past incidents, near misses, tickets, and defects for recurring failure patterns, turning real base rates into likelihood estimates and observed precursors into detection signals for a new design', so its operative form is Analysis, Modeling & Optimization.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Engineering & Design
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Mining past defects and near misses to estimate failure modes is a reliability and safety-engineering practice.
Related originating lineages:
- Data Science & Analytics — Mining heterogeneous records for precursors supplies the analytic implementation.
- Psychology — Base-rate neglect explains why explicit cross-incident review corrects intuitive overreaction to vivid cases.
- Statistics & Experimental Design — Frequency estimation and reference-class sampling provide the inferential machinery.
Review resolution: Both reviewers independently assign engineering_design as the primary originating domain, so that shared primary is retained. Alternate domains are the union of reviewer-identified formative or independently originating lineages; later application settings alone are excluded. The final form materially composes methods or concepts from more than one formative domain. It has established independent use across several domains, but that does not make it domain-free. The encyclopedia entry makes that composition explicit.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; medium confidence.
Notes¶
The value of the corpus rises sharply when it includes near misses, not just incidents that caused harm — they are the same failure paths caught one layer earlier, and there are many more of them. A review that mines only the events serious enough to have hurt someone is working from the smallest and most biased slice of its own evidence.
[n1] A base rate is the underlying frequency of an event in a reference population; base-rate neglect, documented by Kahneman and Tversky, is the well-known tendency to ignore that frequency in favor of a vivid specific case. Incident pattern review is the corrective: it forces the recorded frequency back into an estimate that intuition would otherwise skew. ↩