Cynefin: The Map I Wish I Had Before Every Hard Decision
I applied one playbook — analyze, plan, execute — to every problem. Cynefin named why that failed: half my problems weren't the kind planning can solve.
Cynefin is a sense-making framework from Dave Snowden (Welsh: roughly “habitat”). Its claim:
Different problems need different methods. Most mistakes come from using the wrong domain’s method.
I kept running analyze → plan → execute on everything. Half my problems weren’t that kind.
Two Cuts
Not a Cartesian matrix — Snowden is explicit about that. I still read it as two cuts:
- Ordered ↔ Unordered — Can cause and effect be established in principle?
- A second cut that depends on the side — not one shared Y-axis.
flowchart LR
subgraph unordered ["Unordered — not knowable in advance"]
direction TB
CX["<b>Complex</b><br/><i>enabling constraints</i><br/>probe → sense → respond"]
CH["<b>Chaotic</b><br/><i>absent constraints</i><br/>act → sense → respond"]
CX --- CH
end
subgraph ordered ["Ordered — knowable in principle"]
direction TB
CO["<b>Complicated</b><br/><i>governing constraints</i><br/>sense → analyze → respond"]
CL["<b>Clear</b><br/><i>fixed constraints</i><br/>sense → categorize → respond"]
CO --- CL
end
unordered ~~~ ordered
classDef complex fill:#fff,stroke:#0277bd,stroke-width:2px,color:#000;
classDef complicated fill:#fff,stroke:#f57c00,stroke-width:2px,color:#000;
classDef clear fill:#fff,stroke:#2e7d32,stroke-width:2px,color:#000;
classDef chaos fill:#fff,stroke:#906,stroke-width:3px,color:#000;
class CX complex;
class CO complicated;
class CL clear;
class CH chaos;
- Ordered: obvious → Clear. Needs expertise → Complicated.
- Unordered: patterns can emerge → Complex. Turbulence → Chaotic.
- Disorder: can’t tell which cell — split the problem, sort each piece.
The asymmetry is the point. A shared Y-axis invents a fake 2×2.
| Domain | Constraints | Cause & effect | First move |
|---|---|---|---|
| Clear | Fixed | Known | Categorize |
| Complicated | Governing | Knowable | Analyze |
| Complex | Enabling | Hindsight only | Probe |
| Chaotic | Absent | Not perceivable | Act |
Match the method to the constraint regime. Don’t import a neighbor’s playbook.
The Domains
Clear
Cause and effect are obvious. Missing config? Fix it, write the runbook.
Trap: complacency, not difficulty. Last year’s checklist fails quietly when something underneath changes. Audit and retire rules that outlived their reason.
Complicated
Cause and effect exist, but need analysis — performance tuning, schema design, migrations. Sense → analyze → respond: RFCs, RCA, people who know the domain.
Trap: analysis past the point of value. I once spent three days on a load-balancing choice where any reasonable option would have worked.
Complex
Cause and effect only in hindsight — people, teams, markets, codebases interacting. Flaky monoliths. Morale. Which feature users adopt.
Move: probe → sense → respond. Feature flags, A/B tests, one-sprint tries. Not lazy planning — the correct method. Planning harder is the incorrect one.
Two mistakes: plan a complex problem as if analysis could freeze it; or “try three architectures” when analysis would have answered. Over-planning the unknown usually costs more.
Chaotic
Prod is down. No time to analyze or probe. Act → sense → respond: stabilize first, diagnose second. Chaos is usually complex or complicated that got ignored until it tipped.
Disorder
Not a territory — the state of not knowing which domain. Exit: split the problem, sort each piece.
The Complacency Loop
Domains drift.
flowchart LR
CX["<b>Complex</b><br/><i>probe & learn</i>"]
C["<b>Complicated</b><br/><i>analyze & decide</i>"]
CL["<b>Clear</b><br/><i>follow the runbook</i>"]
CH["<b>Chaotic</b><br/><i>stabilize first</i>"]
CX <-->|"probe, learn, codify"| C
C -->|"success breeds habit"| CL
CL -->|"checklist stops matching reality"| CH
CH -->|"stabilize; constraints return"| CX
classDef complex fill:#fff,stroke:#0277bd,stroke-width:2px,color:#000;
classDef clear fill:#fff,stroke:#2e7d32,stroke-width:2px,color:#000;
classDef chaos fill:#fff,stroke:#906,stroke-width:3px,color:#000;
class CX,C complex;
class CL clear;
class CH chaos;
Healthy drift: complex → complicated → clear (experiments become runbooks). The cost: standardization dulls change detection. When the world moves and the runbook doesn’t, Clear collapses into Chaos.
Recovery lands in Complex first — stabilize, then probe — not expert analysis. Treating the aftermath like a solved engineering problem re-enters the wrong playbook.
Keep a channel for probing: small experiments, slack, people who notice when the map stops matching the terrain.
How I Use It
Before I commit to an approach:
- Cause and effect obvious now? → Clear. Checklist + later audit.
- Analysis could find it? → Complicated. Analyze, decide, stop.
- Only knowable in hindsight? → Complex. Cheapest safe-to-fail probe.
- No time? → Chaotic. Act, stabilize, reflect later.
- Can’t tell? → Disorder. Split and sort.
#3 is the one that pays. Debating which option is “right” when users will vote with their feet? Ask what’s the cheapest experiment that teaches us.
Practical Next Steps
- Sort one stuck decision into a domain. Notice which playbook you were about to use.
- If complex: replace the plan with the smallest safe-to-fail probe.
- Audit checklists quarterly. Clear-domain rules rot silently.
- Read Snowden & Boone, A Leader’s Framework for Decision Making (HBR, 2007) — this summary flattens a richer model.
The map doesn’t decide for me. It stops the wrong method — analyzing a crisis like a design doc, or improvising through a problem that needed analysis.
