Inner Loop Latency
The time from changing a line to observing its effect, repeated dozens of times a day and therefore the highest-leverage number in developer experience.
The inner loop is edit, build, run, observe. An engineer traverses it perhaps fifty times a day. At ten seconds that is eight minutes; at three minutes it is two and a half hours, and — more damaging than the arithmetic — it crosses the threshold where attention leaves the problem.
The number is worth measuring because intuition about it is poor. Teams routinely invest in the outer loop, where the pipeline duration is visible on a dashboard, while the inner loop degrades silently as the local build accretes steps and the application acquires dependencies that cannot run locally.
The characteristic degradation in a microservice estate is that running one service usefully requires running eleven. The answers are ranked: make the service runnable in isolation against contract-verified stubs; failing that, provide a shared or remote development environment the local process can attach to. Requiring the whole estate on a laptop is the outcome to design away from, and it is usually reached by default rather than by decision.