A feedback loop is easy to define and surprisingly hard to see. The definition fits on a card: an output that returns, in some form, to influence the input that produced it. The difficulty is that the return path is rarely labelled. It runs through habits, prices, moods, queues and the small decisions of people who never think of themselves as part of a loop at all.
When a loop is visible, it tends to be because somebody built it on purpose. A thermostat is the classic case: a sensor, a set point, a switch. The circuit is drawn on a diagram and the diagram is true. Most loops in ordinary life are not like that. Nobody designed them, so nobody drew them, so the first evidence that they exist is a pattern that keeps repeating after everyone agreed it would stop.
The signature of a loop
There is a useful habit for finding them. Instead of asking what caused an event, ask what the event caused next, and whether any of that found its way back. A queue that grows makes people arrive earlier to beat it, which makes the queue grow sooner. A shortage that is announced makes people buy ahead, which deepens the shortage it described. The event and its consequence are the same shape, offset in time.
The first evidence that a loop exists is a pattern that keeps repeating after everyone agreed it would stop.
That offset matters more than it first appears. Loops with short delays feel like simple cause and effect. Loops with long delays feel like coincidence, or bad luck, or somebody else's fault.1 The longer the delay, the more likely the loop is to be mistaken for a series of unrelated events, each with its own tidy explanation.
Two kinds, one question
It helps to sort loops into two families. Reinforcing loops amplify: more leads to more, less leads to less. Balancing loops correct: they push a quantity back towards a target, even when nobody has written that target down. Most interesting behaviour comes from the two kinds meeting, one accelerating and the other pushing back, with delays deciding which wins and for how long.
The practical question is rarely which family a loop belongs to. It is where the loop can be interrupted without making things worse. Interrupting a balancing loop removes a correction that was quietly doing useful work. Interrupting a reinforcing loop at the wrong point simply moves the pressure somewhere less visible.2
Watching twice
The method, in the end, is patience. A single observation shows an event. A second observation, after the same conditions return, shows whether the system has a memory. Loops live in that memory. They are the reason a system behaves differently the second time, and the reason the second time is usually more informative than the first.