I agree that it would be difficult to build a tool that could guarantee the code will never be called. Instead, I want to build a tool that can allow developers to home in on good candidates for cleanup.

I am proposing to monitor code as it runs in production. This will allow us to be confident in which code is actually getting called, and the inverse: which code is called infrequently. Then we could install probes in code that is suspected to be dead. e.g. explicitly log when the code is called. Then after an appropriately generous period of time (a month or two), if the code still has yet to be called, then we can be more confident when removing it.