---
id: wiki-2026-0508-systems-thinking
title: Systems Thinking
category: 10_Wiki/Topics
status: verified
canonical_id: self
aliases: [Systems Theory, Holistic Thinking, Feedback Loops]
duplicate_of: none
source_trust_level: A
confidence_score: 0.9
verification_status: applied
tags: [systems, thinking, modeling, complexity]
raw_sources: []
last_reinforced: 2026-05-10
github_commit: pending
tech_stack:
  language: methodology
  framework: causal-loop
---

# Systems Thinking

## 매 한 줄
> **"매 part 의 sum 의 X — 매 interaction 의 emergent behavior"**. 매 element 의 isolated analysis 의 대신 매 stocks, flows, feedback loops, delays 의 holistic 의 see. Donella Meadows 의 "Thinking in Systems" 의 canonical — 2026 의 software, climate, policy 의 적용.

## 매 핵심

### 매 핵심 vocabulary
- **Stock**: 매 accumulation (inventory, $, customers, tech debt).
- **Flow**: 매 rate of change (sales/day, hires/month).
- **Feedback loop**: 매 output 의 input 의 영향.
  - **Reinforcing (R)**: amplifies — 매 viral growth.
  - **Balancing (B)**: stabilizes — 매 thermostat.
- **Delay**: 매 cause → effect 의 lag — oscillation 의 cause.

### 매 leverage points (Meadows, ranked)
1. Paradigm (mindset).
2. Goals of system.
3. Self-organization rules.
4. Information flows.
5. Rules / incentives.
6. Negative feedback strength.
7. Positive feedback strength.
8. Material flows / stocks.
9. Numbers / parameters (lowest leverage).

### 매 archetypes
- **Limits to Growth**: R + B → S-curve.
- **Shifting the Burden**: short-term fix 의 weakens long-term solution.
- **Tragedy of the Commons**: shared resource 의 overuse.
- **Fixes That Fail**: 매 quick fix 의 root cause 의 worsen.
- **Success to the Successful**: rich-get-richer.
- **Escalation**: arms race.

### 매 응용
1. Software incident analysis (alerts → fatigue → ignore).
2. Tech debt dynamics (speed ↔ debt ↔ slowdown).
3. Org design (incentives → behavior → outcomes).
4. Climate / policy modeling.

## 💻 패턴

### Causal Loop Diagram (CLD) — text format
```
Tech debt:
  velocity (-) → tech_debt        // less time to fix → debt grows
  tech_debt (-) → velocity        // more debt → slower work
  // R loop: vicious cycle
  
  tech_debt (+) → cleanup_priority
  cleanup_priority (-) → tech_debt
  // B loop: self-correcting (if priority actually given)
```

### Stock-flow — Python (simple SIR)
```python
import numpy as np
import matplotlib.pyplot as plt

def sir(beta=0.3, gamma=0.1, S0=999, I0=1, R0=0, days=160, dt=1):
    S, I, R = [S0], [I0], [R0]
    N = S0 + I0 + R0
    for _ in range(int(days/dt)):
        dS = -beta * S[-1] * I[-1] / N
        dI = beta * S[-1] * I[-1] / N - gamma * I[-1]
        dR = gamma * I[-1]
        S.append(S[-1] + dS*dt)
        I.append(I[-1] + dI*dt)
        R.append(R[-1] + dR*dt)
    return S, I, R

S, I, R = sir()
plt.plot(I, label='Infected')
plt.show()
```

### Limits to Growth — code
```python
def growth_with_limit(K=1000, r=0.1, P0=10, T=200):
    P = [P0]
    for _ in range(T):
        # R loop: r*P (reinforcing)
        # B loop: (1 - P/K) (balancing as P → K)
        dP = r * P[-1] * (1 - P[-1] / K)
        P.append(P[-1] + dP)
    return P
# Logistic curve: explosive then plateau
```

### Behavior over time graph (BoT)
```
^                          ___
|                       __/
|                    __/         ← logistic (limits to growth)
|                ___/
|            ___/
|        ___/
|     __/
|  __/
| /
+----------------------------> time
```

### Archetype detector — checklist
```yaml
shifting_the_burden:
  symptoms:
    - "Quick fix repeatedly applied"
    - "Underlying problem persists or worsens"
    - "Capability for fundamental fix atrophies"
  examples:
    - Painkillers vs. cause
    - Hotfixes vs. refactoring
    - Hiring more on-call vs. reducing alerts
```

### Causal vs. correlational
```python
# Bad: regression on snapshot
# Good: simulate stocks/flows over time, validate
#       with both reference modes (BoT) and structure
```

## 매 결정 기준
| 상황 | Approach |
|---|---|
| Persistent recurring issue | Identify archetype |
| Counter-intuitive outcome | Look for delays / loops |
| Quick decision | Mental CLD sketch |
| Forecast / policy test | Stock-flow simulation |
| Single-cause obvious | 매 systems thinking 의 overkill |

**기본값**: 매 CLD sketch 의 first — 매 archetype 의 명확 시 의 simulation.

## 🔗 Graph
- 부모: [[Cybernetics Foundations|Cybernetics]]
- 응용: [[Tech Debt]]
- Adjacent: [[Feedback Loops]] · [[Mental_Models|Mental Models]] · [[Causal Inference]]

## 🤖 LLM 활용
**언제**: archetype identification, CLD draft from narrative, 매 hidden loops 의 surface.
**언제 X**: quantitative simulation (use Vensim, Stella, SimPy) — LLM 의 numeric simulation 의 unreliable.

## ❌ 안티패턴
- **Linear thinking on systemic problem**: 매 single cause 의 search — 매 loop 의 miss.
- **Ignore delays**: 매 oscillation 의 surprise — 매 delay 의 model.
- **Optimize parts in isolation**: 매 local optim 의 global degrade.
- **Paralysis by complexity**: 매 over-modeling — 매 핵심 archetype 의 enough.

## 🧪 검증 / 중복
- Verified (Donella Meadows "Thinking in Systems"; Senge "5th Discipline"; Sterman "Business Dynamics").
- 신뢰도 A.

## 🕓 Changelog
| 날짜 | 변경 |
|---|---|
| 2026-05-08 | Phase 1 |
| 2026-05-10 | Manual cleanup — stocks/flows, archetypes, CLD, leverage points |