proving-code-is-deadlisted

# Proving code is dead An agent accretes speculative surface fast and evenly: an option nobody sets, an event nobody subscribes to, a validator guarding a boundary that does not exist. Each one looked reasonable when written, and each one now costs a reader's attention forever. This skill owns the evidence standard for a removal. It does not make product decisions: a candidate with a live production caller is a feature change, and that is somebody's call rather than a cleanup. Removing it is only safe if "nothing uses this" is **proven** rather than grepped-at. So the standard here is high on purpose: prefer a few well-proven candidates over a pile of thin guesses. Read `protected_seams`, `source.production`, and `source.non_production` from `.ledger.yml` before you start. The seams list is what stops a confident wrong deletion. ## What counts as a strong candidate A strong candidate removes, folds, or demotes something real, with clear evidence that the current design costs more than it buys: - A public method, event, option, notification, helper, package, durable record, or test artifact has **no production consumer**. - Tests or docs are the only consumers, and the behavior they pin is not load-bearing. - Two representations mirror the same fact. - An interface has methods every implementation must support and no consumer calls. - A separate package exists only for test, demo, or support code, and adds publish or dependency overhead. - A feature implements speculati