| name | cynefin-classification |
| description | Classifies a decision situation into one of Cynefin's five contexts — Clear, Complicated, Complex, Chaotic or Disorder — and names the matching decision mode, so recommendations are framed as playbooks, expert analysis, safe-to-fail probes or stabilising action. Use before advising action under uncertainty — "classify this with Cynefin", "is there a best practice here or is this new territory?", "what kind of problem is this — known, knowable, or emergent?", "nobody agrees which kind of problem this even is". Not for enumerating a chosen plan's failure modes (use `premortem-analysis`) or designing the probe itself (use `cheapest-experiment`). |
| license | MIT |
| metadata | {"category":"decision-strategy","method":"Cynefin framework","origin":"David J. Snowden and Mary E. Boone, 2007 (Harvard Business Review); Cynefin Co renames, 2020","version":"2.0.0"} |
Cynefin Classification
Cynefin sorts a situation by its kind of cause-and-effect relationship, because each kind rewards a different way of deciding — applying the wrong one is how competent leaders fail in unfamiliar territory. Snowden and Boone set out the framework in "A Leader's Framework for Decision Making" (Harvard Business Review, 2007), naming five contexts: Clear (called Simple in 2007, renamed by the Cynefin Co in 2020), Complicated, Complex, Chaotic, and Disorder — where it is unclear which of the other four applies, renamed Confused / Aporetic (A/C) in 2020. It prevents treating a Complex situation as Complicated: hiring an expert to "analyse and answer" a question whose causation is only visible in hindsight, dressing untested hypotheses as best practice.
When to invoke
Invoke when:
- Action is about to be recommended in an unfamiliar, fast-moving or contested space — new regulation, emerging technology, a novel business model — and the framing (answer vs experiment) is unsettled.
- Someone asks "what's the best practice here?", "should we copy what worked at a peer?", or "how do we navigate this?" in a domain where no playbook may exist.
- Recommendations are in fact probes but are written as commitments, or a review found a Complex problem handled as a Complicated one.
Do NOT invoke when:
- The brief is purely descriptive — for industry structure use
five-forces-analysis, for a market map position-competitor.
- A course of action is already chosen and its failure modes are the question — use
premortem-analysis.
- The domain is settled and what is needed is the probe or pilot design itself — use
cheapest-experiment.
- The question is how an environment could evolve over years rather than what kind of problem it is now — use
scenario-planning.
Procedure — classify, name the mode, state the implication
1 — Classify the context
Work the diagnostic questions in order and pick exactly one context:
| Context | Causation | Diagnostic question | Decision mode |
|---|
| Clear (Simple, 2007) | Obvious to everyone; the realm of known knowns | Is the right answer self-evident and undisputed? | sense – categorise – respond |
| Complicated | A clear cause-and-effect relationship exists but not everyone can see it; known unknowns | Can experts discover the answer with analysis and effort? | sense – analyse – respond |
| Complex | Right answers cannot be ferreted out; patterns emerge only in retrospect; unknown unknowns | Will the causation only be legible afterwards? | probe – sense – respond |
| Chaotic | No manageable patterns; relationships shift constantly; unknowables | Is the situation acutely unstable and demanding immediate action? | act – sense – respond |
| Disorder (Confused / Aporetic, 2020) | It is unclear which of the other four applies | Are people arguing past each other from different implicit contexts? | break the situation into its constituent parts and assign each to one of the other four |
Snowden and Boone's rule for Disorder is explicit: "the way out is to break the situation into its constituent parts and assign each to one of the other four realms". Never act from the centre — a decision taken while in Disorder defaults to whatever mode the loudest participant is most comfortable with.
2 — Name the matching decision mode
State the mode, not just the context — the mode changes how the reader consumes the recommendations: Clear → apply the documented playbook; Complicated → commission expert analysis and expect several defensible answers; Complex → run safe-to-fail experiments, amplify what works, dampen what does not; Chaotic → act first to establish order, sense where stability appears, then move the situation towards Complex. Write it as a contrast where the mismatch is live: "the right mode here is probe – sense – respond, not sense – analyse – respond".
3 — Split, if the answer was Disorder
When the classification is Disorder, decompose the situation into named parts and classify each separately, then report the parts and their contexts rather than one label. A situation that cannot be decomposed is still Disorder and the honest output says so, along with what information would resolve it.
4 — State the implication and check for consistency
Write one sentence telling the reader how to consume the rest of the analysis given the context — for Complex, that the recommendations are hypotheses to falsify, each needing a readout date. Then check the recommendations actually match: if the classification says Complex but the text says "the proven approach", either the wording or the classification is wrong. Do not hedge across two contexts; if the choice is genuinely impossible, that is Disorder, which has its own rule.
Output template
**Decision context:** {Clear | Complicated | Complex | Chaotic | Disorder} — {one-line reason keyed to causation}
**Decision mode:** {sense-categorise-respond | sense-analyse-respond | probe-sense-respond | act-sense-respond | decompose-and-assign}
**Implication:** {one sentence on how to read the recommendations that follow}
For Disorder, the decomposition is mandatory and replaces the single-context line:
**Decision context:** Disorder — {why no single context governs}
**Parts:**
- {part 1} → {context} → {mode}
- {part 2} → {context} → {mode}
**Implication:** {how each part is to be handled; what would resolve the remaining ambiguity}
All three fields are mandatory; a context without its decision mode is decorative, and a recommendation set that contradicts the stated mode must be rewritten before it ships.
Worked example
Illustrative case: in March 2027 Meridian Bank, a mid-size European lender with 4,200 staff, must respond to a regulator's new consultation paper on AI in credit decisions. No compliance practice exists yet; three peer banks have taken visibly different positions; two internal pilots produced contradictory results — a 12 % improvement in default prediction on one portfolio and a 3 % degradation on another.
The naive reading is Complicated: hire a Big Four adviser to analyse the paper and produce the answer. The diagnostic says otherwise — no playbook exists, peer responses diverge, and the pilot results only make sense in retrospect, so causation is visible only afterwards.
**Decision context:** Complex — brand-new regulation, no established practice, divergent peer responses, contradictory pilot results
**Decision mode:** probe-sense-respond
**Implication:** the options below are safe-to-fail probes, not a compliance plan; each carries a readout date and a dampening rule, and none should be scaled before its readout.
Calibration contrast: had the question been "implement the controls the regulator finalised in 2025", the context would be Complicated (expert analysis of known rules, sense-analyse-respond); "apply the published control checklist to a standard mortgage product" would be Clear. Had the bank instead been mid-incident with a live model outage across all lending, it would be Chaotic — act first to restore a manual process, then reclassify.
Verification
Before the classification ships:
Pair with adjacent skills
cheapest-experiment — designs the probe once the context says probe-sense-respond; a Complex classification without bounded, safe-to-fail probes is unfinished.
premortem-analysis — assumes a path is chosen; Cynefin asks whether that kind of bet is even the right kind for the context.
key-assumptions-check — surfaces the premises behind a Complicated-domain expert analysis.
three-horizons — the far-horizon bets in a portfolio are usually Complex, near-horizon ones Clear or Complicated.
evolution-stage — early-stage components sit in Complex territory; commodity components are Clear.
Anti-patterns
- Do not classify as Complicated to flatter the analysis. Most emerging-technology and new-regulation briefs are Complex; defaulting to Complicated lets a report sound expert without the honest uncertainty.
- Do not hedge between two contexts. Picking is the work; genuine inability to pick is Disorder, which has its own rule — decompose and assign, never act from the centre.
- Do not name a context without its decision mode; the mode is what changes the reader's behaviour.
- Do not write Complicated-mode language ("the proven approach") under a Complex classification.
- Do not treat a Complex classification as licence for unbounded experiments — probes are small, fast and safe to fail.
Reference
- D. J. Snowden and M. E. Boone, "A Leader's Framework for Decision Making," Harvard Business Review, vol. 85, no. 11, pp. 68–76, Nov. 2007. PMID 18159787. https://hbr.org/2007/11/a-leaders-framework-for-decision-making — the five contexts, their decision modes and the Disorder rule.
- C. F. Kurtz and D. J. Snowden, "The new dynamics of strategy: Sense-making in a complex and complicated world," IBM Systems Journal, vol. 42, no. 3, pp. 462–483, 2003. doi:10.1147/sj.423.0462 — the original Cynefin paper.
- D. Snowden, "Cynefin St David's Day 2020," parts 1 and 2 of 5, The Cynefin Co, Mar. 2020 — part 1 introduces both Obvious→Clear and Disorder→A/C (Aporetic or Confused), https://thecynefin.co/cynefin-st-davids-day-2020-cynefin-framework/; part 2 elaborates the A/C terminology, https://thecynefin.co/cynefin-st-davids-day-2020-2-of-n/.