unsloplisted

# Unslop Find code that should not exist — **dead code** that no longer runs, and **slop** that runs and earns nothing — then prove each candidate before touching it. The product of a run is a ranked report with the evidence attached; removal follows the user's approval, one deletion per commit. ## Two problems, two epistemics Keep them apart at every stage. Auditing them in one pass is the error that produces confident wrong deletions. | | Dead code | Slop | | --- | --- | --- | | Question | Is this **reachable**? | Should this **exist in this form**? | | Decided by | Reference graphs, call graphs, runtime evidence | Reading the code | | Core | Decidable, with documented blind spots | No decidable core | | Terminates in | A removal backed by evidence | A ranked report for a human | **No tool decides deadness.** Each one decides a *decidable approximation* of it, and the identity of that approximation — never the tool's confidence score — determines whether removal is safe. Perfect detection reduces to the halting problem, so every tool is either unsound for deletion or incomplete. **Authorship is not the defect.** No detector of LLM-written source exists, and every pattern in this skill is abundantly present in human-written code. Report the defect and the rule it violates; leave authorship out of the finding. ## Vocabulary - A **suspect** is a raw tool hit or a read observation. It is not yet a finding. - To **exonerate** a suspect is to find the legitimate mechani