Feedback Loops

Feedback Loops

Overview

A system has a feedback loop when its output circles back as input to the next cycle. Reinforcing loops amplify (compound interest, viral growth, bank runs, death spirals). Balancing loops self-correct (thermostats, price discovery, immune response). The critical complication is delay: when delay is long relative to response time, even well-designed balancing loops produce oscillation and overshoot — and operators systematically mismanage the system (Sterman 1989: supply-line underweight = 0.34 on a 0–1 scale).

Composes with: second-order-thinking · s-curve-technology-adoption · prisoners-dilemma · probabilistic-thinking

When to Use

Apply when: system shows non-linear surprise (collapse, oscillation, death spiral, growth flywheel); you are intervening in a complex system and success depends on how it responds; trends are not extrapolating well; bullwhip or oscillation in any quantity that should be steady; a capex/AI-adoption flywheel is compounding and you need to know when the balancing limits (power, supply, cost, AI-native competition) will bite and whether it will overshoot.

When NOT to use: one-shot linear decision with no feedback; insufficient data to map loops (hand-waving without structure); decision too time-bounded for delays to matter; exogenous shock dominates internal dynamics.

Coaching Novices (Adaptive Front Door)

• Engine mode: concrete case → run The Process directly.
• Coach mode: unfamiliar or no concrete case → guide, don't lecture.

In Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.

1. One-line what-it-is: when a system's output circles back as input, you have a feedback loop — it self-amplifies (reinforcing) or self-corrects (balancing), and delays make behavior far worse than expected.
2. Check fit against When to Use / When NOT to use. If it's a one-shot linear decision, redirect.
3. Elicit their real case: a specific behavior or dynamic they face right now — not a hypothetical.

[WAIT — do not advance until user responds]

1. Run The Process one step at a time with their input — map the loop, classify it, locate the delay.

[WAIT — do not advance until user responds]

1. Close by naming the leverage point uncovered and why it is higher than a parameter fix.

[WAIT — do not advance until user responds]

The Process

Run the Feedback-Loop Diagnosis — map structure, find dominant loop, predict behavior, find leverage.

1. Name system + variable of interest. Without a specific variable, analysis becomes vague narrative.
2. List drivers and outputs. What inputs change your variable? What does it change in turn? Stay concrete.
3. Identify loops. Trace chains where a variable feeds back to itself. Most systems have several.
4. Classify each loop (R or B). Count negative signs around the loop — even = reinforcing; odd = balancing.
5. Locate delays. Where does a cause take significant time to produce its effect? Delays are where intuition fails.
6. Identify dominant loop. Growth phase = R dominant; maturity = B catching up; crisis = suppressed R taking over.
7. Map stocks and flows. Stocks = accumulations; flows = rates. A positive flow can still leave a stock dangerously low.
8. Predict behavior pattern. Pure R → exponential growth/collapse. Pure B → equilibrium. R+delay → overshoot/oscillation. R+B competing → S-curve. Mismatch with observed behavior = missed loop.
9. Find leverage (Meadows hierarchy). Parameters → buffers → structures → delays → balancing loops → reinforcing loops → goals → paradigm. Most failed interventions push parameters; move up.
10. Stress-test against system response. Balancing loops fight back; reinforcing loops restore trajectory. Intervention must change structure, not just symptom.

Output: Feedback-Loop Diagnosis

System / variable: <…>
Loops: R1 ; B1 
Delays: 
Dominant loop: <…> — matches observed behavior because <…>
Stocks: <…>  Flows: <…>
Predicted behavior without intervention: 
Leverage (Meadows): lowest ; higher ; highest 
Intervention:  | System response: <…> | Backfire risk: <…>
Falsifier: 

→ Method in Action: Forrester's Beer Distribution Game & Sterman's 1989 Measurement

→ 2026 lens: The AI Capex Boom as a Reinforcing Loop Meeting Its Balancing Limits (2024–2026)

Pack: Loop Patterns

• R growth: network effects, viral k>1, compounding learning curves. Risk: hits a balancing limit you don't control.
• R collapse: death spirals, bank runs, adverse selection cascades. Defense: structural circuit breakers, fast intervention.
• B working: market price discovery, wages, thermostats. Don't suppress healthy balancing loops.
• B + long delay → oscillation: bullwhip, cobweb cycles, capacity build-out. Defense: shorten delays, damp response, share end-demand data.
• R + B → S-curve: technology adoption. See s-curve-technology-adoption. To extend growth, kick off a second R loop before the first saturates.

Common Rationalizations

[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.

Red Flags

• Trends extrapolated in a feedback-driven system · Oscillation blamed on external variability without checking internal loop generators · Intervention at parameter level when loop structure is the source · "Be more careful" proposed in a Forrester-adversarial structure · Stocks and flows confused · Death spiral or growth narrative with no loop/nodes/delays specified · All interventions at lowest (parameter) leverage level

Verification

• System and variable named · At least one R and B loop identified with causal chain · Each loop classified by sign-counting
• Delays identified with rough magnitudes · Dominant loop identified; observed behavior consistent with it
• Stocks and flows distinguished · Predicted behavior matches actual (if not, re-classify)
• Leverage points ranked; recommendation not at lowest level if higher leverage is accessible
• System response to intervention considered · Observable falsifier named

→ Primary sources: references/sources.md

---

Part of deciqAI Knowledge Skills — 227 open-source thinking skills that make rigor executable for AI agents. The same skills power every deciqAI agent, which runs them autonomously to operate your company. See it run → https://www.deciqai.com/c/feedback-loops · ⭐ Star the repo → https://github.com/deciqAI/knowledge-skills · Contributions welcome.

Agents: latest version & machine-readable metadata → https://www.deciqai.com/s/feedback-loops.json