Systems Thinking

Input: $ARGUMENTS

Interpretations

Before executing, identify which interpretation matches the user’s input:

Interpretation 1 — Understand why a system behaves the way it does: The user sees outcomes they did not design or intend and wants to understand the structural dynamics producing them — feedback loops, delays, accumulations. Interpretation 2 — Find leverage points: The user wants to change a system’s behavior and needs to identify where small interventions produce large effects — and where large interventions get absorbed. Interpretation 3 — Anticipate emergent behavior: The user is designing, building, or modifying a system and wants to predict unintended consequences, oscillations, or perverse incentives before they manifest.

If ambiguous, ask: “I can help with understanding why a system behaves as it does, finding leverage points to change it, or anticipating emergent behavior from a design — which fits?” If clear from context, proceed with the matching interpretation.

Core Principles

1. Structure drives behavior. Systems produce outcomes because of their structure — feedback loops, delays, information flows — not because of the intentions of the people in them. If you do not like the behavior, change the structure, not the actors.

2. Stocks change slowly; flows change fast. Stocks (accumulated quantities — trust, inventory, debt, skill) are what give systems memory and inertia. You cannot change a stock instantly. Interventions that target flows instead of stocks get faster results but often miss the deeper dynamic.

3. Delays are invisible causes. When cause and effect are separated in time, people misattribute outcomes. Most policy oscillations — overbuilding then cutting, overhiring then firing — result from acting on delayed feedback as if it were current.

4. Every balancing loop has a goal; every reinforcing loop has a direction. Balancing loops resist change — find the implicit goal they are serving (even if nobody set it deliberately). Reinforcing loops amplify — find what they are amplifying and whether that acceleration is sustainable.

5. The system will compensate. Interventions that push against a balancing loop will be absorbed. The system finds another way to maintain its goal. Effective intervention works WITH the system’s structure, not against it.

6. Boundaries are choices, not facts. Every system boundary you draw includes some things and excludes others. If your model cannot explain the behavior, your boundary is probably wrong — you are excluding a critical feedback loop.

Phase 1: MAP — Identify System Structure

Step 1: Define the System and Its Boundary

SYSTEM: [What system is being analyzed]
OBSERVED BEHAVIOR: [What the system is producing — the phenomenon that prompted analysis]
BOUNDARY: [What is inside the system, what is outside]
TIME HORIZON: [Over what period does the behavior manifest]

[S1] Boundary check: What have you excluded that might contain a feedback loop driving the behavior? If the system behavior cannot be explained by what is inside the boundary, expand it.

Step 2: Identify Stocks and Flows

Stocks are accumulations — things that build up or drain over time. Flows are rates of change.

STOCKS (what accumulates):

[S2] Stock: [name]
  Current level: [high / medium / low / unknown]
  Inflows: [what adds to this stock]
  Outflows: [what drains this stock]
  Response time: [how fast does this stock change]

[S3] Stock: [name]
  Current level: [high / medium / low / unknown]
  Inflows: [what adds to this stock]
  Outflows: [what drains this stock]
  Response time: [how fast does this stock change]

[Continue for all significant stocks]

Check: Are there hidden stocks? Common invisible stocks include trust, technical debt, morale, institutional knowledge, backlog, and goodwill.

Step 3: Map Feedback Loops

For each significant behavior, trace the causal loop:

FEEDBACK LOOPS:

[S4] Loop: [name]
  Type: [Reinforcing (R) / Balancing (B)]
  Mechanism: [A increases → B increases → C increases → A increases (R)]
             [A increases → B increases → C decreases → A decreases (B)]
  Speed: [Fast / Medium / Slow]
  Currently: [Dominant / Active / Dormant]
  Implicit goal (if B): [what level is this loop trying to maintain?]
  Direction (if R): [what is being amplified — growth, collapse, polarization?]

[S5] Loop: [name]
  Type: [R / B]
  Mechanism: [trace the chain]
  Speed: [Fast / Medium / Slow]
  Currently: [Dominant / Active / Dormant]

[Continue for all significant loops]

Rules:

• Count the negative links in each loop. Odd number = balancing. Even (or zero) = reinforcing.
• A system with only reinforcing loops will grow or collapse — look for the missing balancing loop
• A system that is stuck has a dominant balancing loop — find it

Step 4: Identify Delays

DELAYS:

[S6] Delay: [between what cause and what effect]
  Duration: [approximate time lag]
  Consequence: [what behavior does this delay produce — oscillation, overshoot, misattribution?]
  Visibility: [is this delay visible to actors in the system?]

[Continue for all significant delays]

Phase 2: ANALYZE — Understand the Dynamics

Step 5: Explain the Observed Behavior

Using the map from Phase 1, explain WHY the system produces the behavior observed:

BEHAVIOR EXPLANATION:

[S7] The system produces [observed behavior] because:
  1. [Loop X] drives [what] — this is currently [dominant/active/dormant]
  2. [Stock Y] is [high/low/changing] because [inflow/outflow imbalance]
  3. [Delay Z] causes actors to [misattribute/overshoot/oscillate]
  4. [Boundary issue] means [external factor] is not being accounted for

KEY DYNAMIC: [The single most important structural explanation]

Step 6: Identify System Archetypes

Check against common system archetypes:

ARCHETYPE MATCH:

[S8] Archetype: [name]
  How it applies: [specific mapping to this system]
  Predicted trajectory: [what happens next if structure unchanged]
  Classic intervention: [what the archetype literature suggests]

Phase 3: INTERVENE — Find Leverage Points

Step 7: Identify Leverage Points

Rank potential interventions by leverage (Meadows’ hierarchy, adapted):

For each applicable level, provide: the specific intervention, its leverage assessment, and a concrete example in this system.

Step 8: Evaluate Intervention Side Effects

For each proposed intervention, trace it through the system map:

INTERVENTION: [proposed change]

FIRST-ORDER EFFECT: [immediate intended result]
SECOND-ORDER EFFECT: [what that result causes through feedback loops]
THIRD-ORDER EFFECT: [what the second-order effect causes]
COMPENSATING RESPONSE: [how the system's balancing loops will resist this change]
DELAY RISK: [will delayed feedback make this look like it is working when it is not, or vice versa?]

NET ASSESSMENT: [After accounting for system response, will this intervention achieve its goal?]

Failure Modes

Depth Scaling

Default: 2x. Parse depth from $ARGUMENTS if specified (e.g., “/systhink 4x [input]”).

These are floors. Go deeper where insight is dense. Compress where it is not.

Pre-Completion Checklist

• System boundary explicitly defined and checked for missing loops
• All significant stocks identified with inflows and outflows
• Feedback loops traced with type (R/B), speed, and dominance
• Delays mapped and their behavioral consequences stated
• Observed behavior explained by structure, not by events or actors
• At least one system archetype checked for fit
• Leverage points ranked from high to low, not just parameter-level fixes
• Intervention side effects traced through at least two orders of consequence
• Compensating responses from balancing loops anticipated

Integration

• Use from: /rca (when root cause is structural, not event-level), /fohw (understanding how something works as a system), /insd (when inside-view misses structural dynamics)
• Routes to: /prob (when system behavior involves uncertain feedback), /aex (when assumptions about system structure need testing)
• Differs from: /rca traces backward from failure to cause; /systhink maps the ongoing structural dynamics that produce behavior patterns
• Complementary: /fohw (mechanistic understanding), /insd (revealing hidden structure), /reframe (changing which system you are looking at)