Systems Thinking
Reasoning about a system in terms of the interactions and feedback loops between its parts, rather than the parts individually.
Most production surprises are not component failures; they are interactions. A retry policy that is correct in isolation combines with a timeout that is correct in isolation to produce a retry storm. An autoscaler responds to the load the retry storm creates. Each part behaved as designed.
What the habit adds in practice: looking for feedback loops (does this response make the cause better or worse?), for delays (how long between the action and its visible effect, and what will someone do during that gap?), and for the second-order effect of every mitigation.
It is also the source of the most useful review question available: "what does this do when it is overloaded, and does that make the overload better or worse?"