I think it's both attention-getting and distracting to start with a definition of unit testing that hardly anybody uses. Now I'm not interested in the article

I think it's both attention-getting and distracting to start with a definition of unit testing that hardly anybody uses. Now I'm not interested in the article because I have to see what your sources are and whether you're gaslighting me. The reason people use the term unit test to mean the size of the system under test is because that's what it's generally meant. Before OO, it would mean module. Now it means class. The original approach would be to have smaller, testable functions that made up the functionality of the module and test them individually. Decoupling was done so that you didn't need to mock the database or the filesystem, just the logic that you're writing. Some people disagree with unit testing and focus on functional testing. For example, the programming style developed by Harlan Mills at IBM was to specify the units very carefully using formal methods and write to the specification. Then, black-box testing was used to gain confidence in the system as a whole. I feel that a refactor shouldn't break unit tests, at least not if the tools are smart enough. If you rename a method or class, its uses should have been renamed in the unit tests. If you push a method down or up in a hierarchy, a failing test tells you that the test is assuming the wrong class. But most cases of failing tests should be places where you made a mistake. However, I agree that functional tests are the hurdle you should have crossed before shipping code. Use unit testing to get 100ms results as you work, functional tests to verify that everything is working correctly. Write them so that you could confidently push to production whenever they're green.