Integration Frequency
How often each engineer merges to the mainline — the measurement that distinguishes continuous integration from having a CI server.
Continuous integration is a practice about merge cadence, not a category of tooling. The question that settles whether a team is doing it is: how long does the average change live outside the mainline? If the answer is days, the team has a build server and long-lived branches, and it is paying the integration cost it believed it had automated away.
The mechanism is straightforward. Merge conflicts and semantic clashes grow super-linearly with divergence, so daily integration produces small, cheap conflicts and fortnightly integration produces the merge everyone dreads. The dread then justifies branching further, which is the loop.
Integration frequency is also a proxy for several other things: it cannot be high if the test suite is slow, if review takes days, or if the work is not being broken into pieces that are individually safe to ship. That makes it a good leading indicator — when it falls, one of those three has degraded, and it is worth finding out which before lead time follows.