DeadCode - Delete with Evidence

Dead code accumulates in every growing software project: unused functions, classes, imports, legacy modules, and abandoned code paths. The difficult part is not identifying code that appears unused it is proving that removing it will not break the application. DeadCode is a deterministic developer tool designed to solve this problem. DeadCode scans Python repositories using AST-based static analysis and reference/dependency analysis. It identifies suspiciously unused code and classifies candidates into three states: PROVABLE, REVIEW, and ACTIVE. Each candidate is accompanied by structured evidence such as its definition, references, imports, test references, exports, and safety-related signals. The key difference is that DeadCode does not immediately delete a candidate. A PROVABLE candidate can first be passed through the proof engine. DeadCode creates an isolated Git worktree, removes the candidate there, runs available verification checks, and re-runs its analysis to confirm that the candidate has disappeared. Only after successful proof can the developer explicitly apply the removal to the working tree. When DeadCode encounters uncertainty such as framework registration, dynamic behavior, entry points, test discovery, or other safety-sensitive patterns it fails closed and classifies the candidate as REVIEW instead of assuming it is safe. The product is available as a developer CLI and as a browser-based demonstration interface. The live demo allows users to analyze a prepared Python repository or a public Python GitHub repository, inspect candidates, prove removals, and apply verified changes. DeadCode is intentionally deterministic at runtime: no LLM is used to make deletion decisions. The result is a safer and more explainable workflow for reducing codebase complexity: find it, prove it, remove it.