carlheath Systems Thinking
Role: structural analysis of complex systems. Tradition: Meadows / Senge / Forrester / Ackoff / Santa Fe Institute.
Five workflows
1. Iceberg (Iceberg model)
Layers downward: Events → Patterns → Structures → Mental Models.
🌊 What happened? (Events — visible)
↓
📊 How often does it happen? (Patterns — tendencies)
↓
🏗️ What in the structure produced it? (Structures — institutions, incentives, rules)
↓
🧠 What mental models support the structures? (Mental Models — assumptions)
Use for recurring problems: "Why does X keep happening?" The Iceberg reveals events as symptoms; the solution lies in structures or mental models.
2. CausalLoop (Causal Loop Diagrams)
Build diagrams with:
- Reinforcing loops (R) — self-amplifying (positive feedback)
- Balancing loops (B) — self-stabilising (negative feedback)
Output as a mermaid diagram:
graph LR
A[Reliance on social media for news] -->|+| B[Algorithm-driven attention]
B -->|+| C[Polarisation]
C -->|+| A
C -.->|R1| A
3. FindArchetype (10 canonical archetypes)
Match behaviour to known patterns:
- Limits to Growth — growth hits a ceiling
- Shifting the Burden — symptom-fix replaces root action
- Tragedy of the Commons — shared resource overused
- Fixes That Fail — short-term fix worsens the long term
- Eroding Goals — goals lower when things get hard
- Escalation — A's reaction provokes B, and vice versa
- Success to the Successful — winners get more resources
- Drifting Goals — gradual standard erosion
- Growth and Underinvestment — growth without capacity investment
- Accidental Adversaries — partners become opponents via misunderstanding
Identify the archetype → apply the canonical intervention.
4. FindLeverage (Meadows' 12 leverage points)
Ordered weakest → strongest:
- Constants & parameters (weakest)
- Buffer sizes
- Stock-flow structures
- Delays
- Balancing feedback loops
- Reinforcing feedback loops
- Information flows
- Rules
- Self-organisation capacity
- Goals
- Paradigm
- Power to transcend paradigms (strongest)
Application: policy work should aim for layers 5-1 for real impact, not 12-7 (which is what politics often does).
5. ConceptMap (Novak-style entity-relationship)
Hierarchical concept map:
- Overarching concept on top
- More specific downward
- Linking phrases between concepts ("causes", "amplifies", "part of")
Output as mermaid or hierarchical list.
Domain examples
| Question | Workflow | Expected outcome |
|---|---|---|
| "Why does digital transformation keep failing?" | Iceberg | Mental models about "tech = solution" rather than organizational change |
| "Polarisation around social media" | CausalLoop | R-loop: attention → algorithm → polarisation → more attention |
| "Decline of defense spending pre-crisis" | FindArchetype | Eroding Goals (target lowered over decades) |
| "How to influence digital sovereignty in policy?" | FindLeverage | Aim for layers 5-3 (rules, goals, paradigm) not 12-10 (more money) |
Integration with other skills
- thinking-firstprinciples to deconstruct mental models in Iceberg
- thinking-redteam to stress-test interventions from FindLeverage
- root-cause-analysis for incident cause chains (complementary — RCA is linear, systems-thinking is cyclic)
Output format
Mermaid diagrams for CausalLoop + ConceptMap. Structured markdown for Iceberg + FindArchetype + FindLeverage.
🎯 COMPLETED: [SKILL:systems-thinking] [systems analysis of X] 🗣️ CUSTOM COMPLETED: [SKILL:systems-thinking] [Systems analysis done]