Every other kind of thinking assumes something holds still.

Analysis assumes the parts hold still. Systems thinking assumes the structure does. Design thinking assumes the user does. Complexity thinking starts where those assumptions fail: when behaviour comes from interaction, the rules shift as things unfold, and the people analysing are part of what they analyse. In a school, that is most of the time.

The toolkit

Why it matters

Complexity fails in two opposite directions. In technical settings it is reduced to technique: network statistics and agent-based models, with nothing said about judgement. In management and policy it is inflated into rhetoric. “It’s complex.” “We live in a VUCA world.” Both hide what is distinctive about it.

Underneath sits a literacy gap. Jacobson and Wilensky found that students default to single-cause, centralised explanations of patterns that are in fact emergent. Leaders default to plan and control. Institutions default to linear accountability. The world has become more complex than the mental and institutional models we use to run it.

And education is now at the centre of the problem, not its edge. Rahwan and colleagues describe algorithms acting as agents inside social systems, producing behaviour nobody intended. Generative AI brings that into knowledge, learning and communication, which is to say, into schools.

What I have come to believe

It is a diagnostic first. Complexity thinking is needed when three conditions hold together. Behaviour comes from interaction, which is emergence. The rules change as things unfold, which is adaptation. Our framing and action feed back into the situation, which is reflexivity. Each condition alone belongs to an existing discipline. All three together are complexity’s home ground. Where they do not all hold, other thinking serves better. A bridge is complicated, not complex. A payroll migration is a technical problem, not an adaptive one. Using complexity there is a category error.

Systems thinking maps structure. Complexity asks how structure appears and shifts. That is a difference of starting assumptions, not of tools. The practical test is one question: is this a system you can design, or one you can only steer from inside?

It works best as an overlay, not a replacement. Seven checks, one for each of the seven contributions, can be run on any strategy, risk register, theory of change, dashboard or business case. Each one changes it materially. And robustness replaces precision. The question stops being “is this analysis right?” and becomes “is this robust enough to act on, and safe enough to be wrong about?”

It must meet its own standards: provisional, plural, reflexive and testable. A complexity practice that cannot say what would prove it wrong has become the thing it criticises.

How I approach it

I profile the situation before choosing a method. I run the three-condition test, then set six dials in place of a VUCA label. Most plans contain parts that are merely complicated. Those get planned properly.

I ask what the situation is already inclined to do before asking what we want it to do. Then I commit to a direction, not a destination, and set the triggers for revisiting it in advance. That is provisional commitment.

And I use the tools the team already has. I run the overlay on their risk register or their theory of change instead of introducing a new framework. New frameworks are how complexity becomes a workshop.

The moves

Four moves that carry seven contributions

The other strands have four moves, and so does this one. Each move carries one or more of the seven contributions, so the four cover all seven. Each card opens the move’s own page, set out to try.

Decides which kind of thinking fits

Profiling the situation

Run the three-condition test, name the kind of not-knowing, and set six dials in place of a VUCA label. Most plans are part complicated and part complex. The first job is to stop running both on one logic.

Carries: The diagnostic

Finds what the situation is already inclined to do

Reading the disposition

Map the relationships, not just the actors. Then ask where the system tends to settle and what is changing slowly underneath. Small, well-timed moves matter near a threshold. Large pushes are wasted where it is stable.

Carries: Relations, Dispositions

Learns by acting, without spending the future

Probing with options open

Shape conditions instead of specifying outcomes. Run several safe-to-fail probes, with criteria for amplifying or dampening them set in advance. Audit which options the plan spends, and decide which doors are one-way.

Carries: Conditions, Optionality

Keeps the analysis honest

Holding the frame lightly

Use more than one model and trust what survives across them. Expect people to respond to your measures. Name who and what the frame leaves out, and who pays for it.

Carries: Plural models, The analyst inside, Responsibility

The four moves at a glance, with every worksheet

Limits

Where it stops being useful

Four of the eight ways complexity thinking goes wrong. The other four, with a guardrail for each, are on the test page.

When “it’s complex” becomes an alibi, used to avoid commitment or accountability. The guardrail is to commit to a direction, to learning milestones and to explicit triggers.
When it is applied to the wrong kind of problem. Complexity methods on a clear or complicated problem waste effort and blur accountability. Run the test first.
When options are kept open forever. Optionality becomes indecision. Some problems need deliberate lock-in, because delay is the larger risk.
When everything is connected. If everything is connected, nothing can be prioritised. Most things are weakly connected most of the time. Find the strong couplings and act on those.

Worked example

Complexity, running

The strand taken out and used. Start with the test. The seven contributions are each followed through one school term, as the same school changes how it assesses, of learning and for it. The people are illustrative.

When it is needed, and when it is not

Three conditions, all at once. What each alone belongs to, the thirteen neighbouring ways of thinking, and eight ways complexity thinking goes wrong.

Seven contributions, one arc

What complexity thinking adds that no neighbour does, from what exists to what we owe. Each one followed through the same school across a semester.

The overlay: seven checks for any tool

Keep the strategic plan, the risk register and the dashboard. Run seven checks over them. The school’s assessment plan, before and after.

Compare the contributions

Hold one thing steady (a person, a dimension, the drawing) and let the seven contributions vary.

Reference

The reference shelf

The ideas the strand draws on, set out at length: where the vocabulary came from, what it claims, and where it is weak. Read them when a page above sends you, or when a word needs pinning down.

What I would still argue about

The honest list, and the part of this strand I would defend least confidently.

Two complexities. Rigorous complexity science and the looser complexity thinking used in leadership and futures barely talk to each other. This strand sits on the looser side. I am not sure it has earned the right to borrow the other side’s authority.
The evidence for the tools. Cynefin, Design Unbound and most mapping methods have many case illustrations and few controlled evaluations. The field is open to the charge of being persuasive rather than proven, and so is this page.
Normative thinness. Complexity explains how patterns arise. It says nothing about which ones are desirable. Responsibility for framing tries to supply that, and I do not think it fully succeeds.
The narrow canon. The field is overwhelmingly Western. The traditions page tries to widen it without romanticising or extracting, and that is not a solved problem.
How to evaluate any of this credibly. The outcome I care about is a decision made differently months later. That is the toolkit’s own problem, in a harder form.

Singapore