What a feedback loop is
A feedback loop exists when a system’s output influences its own future input. The cause-and-effect chain goes around in a circle: A affects B, B affects C, and C affects A. Once you notice a circle like this, the usual straight-line question “what caused this?” becomes harder to answer, because everything in the loop is both a cause and an effect.
Feedback is not a special kind of equipment. It appears in a kettle, a team’s bug backlog, a power supply and a public queue. Learning to recognise two basic types, reinforcing and balancing, is enough to make sense of many behaviours that look puzzling at first.
Balancing loops
A balancing loop pushes a system toward a target or limit. When the system moves away from the target, the loop acts to bring it back. Another name is negative feedback, where “negative” describes the direction of the correction, not its quality.
The home thermostat is the standard example. If the room is colder than the set point, the heating turns on; as the room warms, the difference shrinks, and the heating turns off. The loop works to close a gap.
Other examples:
- Cruise control: if the vehicle slows on a hill, the controller asks for more power, which reduces the speed error.
- Voltage regulation: if an output voltage drops, a regulator circuit adjusts to bring it back toward its target.
- Workload on a team: when a queue of tasks grows, people work harder or ask for help, which tends to shrink the queue again.
Balancing loops are the basis of control and of stability, but they are not magic. They have limits, and two features deserve particular care.
Delay
If the correction takes effect slowly, the system can overshoot. Imagine adjusting a shower where the water takes several seconds to respond: you turn the tap, nothing seems to change, you turn it further, and then the water is too hot. The loop is still balancing, but the delay makes it swing back and forth. Similar effects appear in heating systems, in inventory ordering and in how a team reacts to a rising defect count that was measured weeks earlier.
Gain and limits
If the loop reacts too strongly, it can overcorrect and oscillate; if it reacts too weakly, it responds sluggishly. Real devices also have limits, such as a heater that cannot deliver more than its rated power. When the correction needed exceeds what the device can do, the loop cannot hold the target.
Reinforcing loops
A reinforcing loop amplifies change. A small increase leads to a further increase, or a small decrease to a further decrease. Left alone, it grows or shrinks at an accelerating rate until something outside the loop stops it. This is often called positive feedback.
Physical examples:
- Audio howl: a microphone picks up sound from a nearby speaker, the amplifier makes it louder, and the speaker feeds more sound back to the microphone.
- Thermal runaway in some electronic devices: in certain conditions, higher temperature leads to higher current, which produces more heat. Whether it occurs depends on the specific device and circuit, which is one reason to respect design margins and the information on a datasheet.
- Snowball effects in a process: a leak lowers pressure, the system compensates by working harder, and the extra stress makes the leak grow. This is an invented illustration, but the pattern is real in many settings.
Organisational examples:
- Rushed fixes: a backlog of defects creates pressure, the team takes shortcuts, the shortcuts introduce new defects, and the backlog grows. Each pass around the loop makes the next one worse.
- Good habits: a team that writes clear documentation finds problems faster, saves time, and so has more time to document. Reinforcing loops can work in a helpful direction too.
Putting both together
Most real systems contain several loops at once. A balancing loop may hold a variable steady until a reinforcing loop grows large enough to overwhelm it. A useful habit is to ask of any recurring behaviour: where is the circle, is it amplifying or correcting, and what limits it?
A short comparison:
| Reinforcing | Balancing | |
|---|---|---|
| Tendency | Amplifies change | Resists change toward a target |
| Typical behaviour | Growth, collapse, runaway | Settling, holding, oscillating if delayed |
| Needs an outside limit | Yes | Usually has its own target |
How to find loops in a design
- List the key variables, such as temperature, queue length, pressure, cost or morale.
- For each pair, ask whether a rise in one tends to push the other up or down.
- Look for chains that return to where they started.
- Mark each link as “same direction” or “opposite direction”. A loop with an even number of opposite links, including zero, tends to reinforce. A loop with an odd number tends to balance.
- Look for delays between links. Long delays are a common source of unexpected oscillation.
As a check on the rule in step 4: in the thermostat, a larger gap leads to more heating (same direction), more heating raises room temperature (same direction), and a higher room temperature reduces the gap (opposite direction). That is one opposite link, an odd number, so the loop balances, as expected.
Practical implications
- Do not treat the symptom alone. If a reinforcing loop drives a problem, repeated local fixes may only slow it. Look for the point where the circle can be broken.
- Add limits to runaway paths. Current limits, thermal protection and queue caps are examples of stopping a reinforcing loop before it reaches harm. Hazardous cases need qualified engineering judgment and the standards that apply in your industry.
- Be careful with delayed measurements. If the thing you measure lags the thing you control, expect overshoot, and consider slower, gentler corrections.
- Check the human parts. Metrics and targets can create loops with people’s behaviour, sometimes in directions nobody intended.
This article is general background. Control design for real equipment involves analysis well beyond these qualitative ideas, and outcomes depend on the specific system.