[marvin]

I'm convinced, after working on a (micro)service-heavy 2+ million line code base for a couple of years now.

Our thousands of unit tests are invaluable, primarily for providing a strict definition of what the code should do, but also for catching bugs when refactoring. We don't often introduce bugs that break tests when refactoring simple parts of the system. But for the complex parts of the system, the tests both maintain a ground truth for how the code should behave, and make it trivial to ensure correct behavior after refactoring.

If we hadn't had the tests, we would be almost guaranteed to change the intended behavior of the code when modifying complex parts of the system. Perhaps no big deal if you're writing an Instagram clone and crashes will show up in your logs and prompt you to investigate, but critical when your code processes millions in financial transactions and subtle logging errors could cause a nightmare.

As always, it depends on your problem domain and priorities. There's an expensive overhead to having good test coverage. Many failed tests will be false positives.