Code Coliseum

Code Coliseum is an evidence-driven arena for AI-generated code. Instead of sending a feature request to one coding agent and trusting the first answer, Code Coliseum launches three IBM Bob contenders with different engineering strategies: Minimalist, Sprinter, and Architect. Each agent receives the same request and works independently inside an isolated Git worktree, so every contender starts from the same codebase and cannot interfere with the others. After implementation, every contender enters a deterministic Gauntlet. The system verifies dependency integrity, TypeScript safety, production build success, and required runtime behavior. Any contender that fails a mandatory check is eliminated. Survivors are compared using real Git evidence such as files changed, lines changed, execution time, and the exact unified diff. The developer stays in control. Code Coliseum does not automatically commit or push AI-generated code. Developers can inspect each surviving implementation, select the candidate that best fits their needs, and apply that exact diff to the working tree for final review. Built with IBM Bob, Next.js, React, TypeScript, Node.js, Git worktrees, and native child-process tooling, Code Coliseum turns AI coding from “generate and hope” into “compete, prove, and merge.”