Dead Code Is Not Just a Cleanup Problem
Every old codebase has it. Methods nobody calls. Endpoints nobody hits. Feature flags whose "off" branch was last evaluated in 2019. Config keys that point at services that were turned off two reorgs ago.
Most teams treat this as housekeeping. It is not. It is a tax on every decision that comes after.
Dead code is not the absence of work. It is the presence of confusion.
The hidden weight of code that does nothing
Dead code does not sit quietly. It shapes how engineers read the system. A new developer opens a class with twelve methods and assumes all twelve matter. They trace dependencies through code paths that have not run in years. They read defensive error handling for failure modes that can no longer happen. They factor a "rare" branch into their design when it is not rare — it is extinct.
Multiply that across a few thousand files and the cost is enormous. Not in CPU. In cognition. Reviews take longer. Refactors get more conservative. Estimates inflate. Onboarding stretches from weeks to months. None of this shows up on a dashboard. All of it shows up in velocity.
It also hides the real dependencies
When live and dead code sit side by side with no visible difference, the actual shape of the system gets blurred. An engineer planning a migration cannot tell which adapters still carry traffic. A security review cannot tell which authentication paths are reachable. An architect drawing the domain map keeps a service in the diagram because the code is still there, even though no one has called it in eighteen months.
Dead code also expands attack surface. Unused endpoints still respond. Old libraries still ship. Forgotten admin routes still authenticate.
The shift
Static analysis is not enough. A linter sees what the compiler sees: references. It cannot tell you whether anything actually called that method last quarter. It cannot tell you which API has not received a request in a year.
The only honest signal of "alive" is production usage. When teams can see which methods, endpoints, and flows actually executed in production — and which have gone silent — dead code stops being a matter of opinion. It becomes a matter of evidence.
Cleanup is easy when you can prove what is dormant.
The point is not to delete more code. The point is to read the system you actually have, not the one the repository remembers.