How Do You Safely Refactor a Large, Untested Legacy Codebase?

Start by creating seams around the code you need to change. Characterize one narrow behavior with an approval or integration test, then put wrappers around the global and external dependencies. After that, make small behavior-preserving changes and run the tests after every step. Don’t try to clean up the entire God object just because you found it; improve the parts you actually need to touch.

The strangler approach works well here: define a small responsibility, implement the replacement beside the old code, and route that responsibility to the new component once its behavior is covered. Repeat this gradually until the old implementation has little left to do. It’s less risky than a rewrite and gives you a clean boundary to work toward.

Your approval tests are a strong starting point. Keep adding focused characterization tests around behaviors as you encounter them, and commit very small changes. Rename things, isolate dependencies, introduce state objects, and extract only when the tests give you confidence. The goal isn’t to make the whole codebase beautiful in one pass; it’s to leave each area safer and easier to change than you found it.