Context Window Stress Test

Evaluates how well tools maintain accuracy when working with large codebases that exceed typical context windows.

Methodology

Each tool was tasked with making 5 specific changes across a 50,000-line TypeScript monorepo with 200+ files. Changes required understanding cross-module dependencies, shared types, and configuration files. Scored on changes correctly made, broken imports, missing updates, and whether the project compiles after changes.