A spy is a wrapper around a real method that already exists. jest.spyOn(obj, 'method') finds obj.method, wraps it in a mock function, and returns that mock, yet behind the wrapper the live implementation is still there. A spy does two things at once: it records every call - the arguments and the number of invocations - and by default it still calls the original and returns its result. So without a single extra line it replaces nothing; it merely steps in between the code and the method to watch it.

This is what sets it apart from jest.fn. jest.fn() is a standalone stub: there is no real implementation behind it, and it does exactly what you told it to. A spy, by contrast, is always tied to something that already exists - a method of a concrete object - and it observes or overrides that very method. Need a new stand-in call that does not exist in the code yet, say a dependency for a service - reach for jest.fn. Need to trace or temporarily change a method of an existing object without rewriting it - that is a job for spyOn.

The most underrated mode is to spy while replacing nothing. Take a CheckoutService with an applyDiscount method that the service calls inside checkout. We put a spy on service.applyDiscount and run checkout as usual. The real discount is computed for real - the spy did not switch it off - and we additionally confirm that the method received the right items and that the total worked out.

The moment you add .mockImplementation or .mockReturnValue, the original is switched off - the spy stops calling the live method and answers by your script. You need this when the real implementation gets in the way: it goes over the network, reads the clock, touches the file system. The classic example is Date.now. The service stamps a receipt with the current time, and without a replacement the test would depend on the second it ran. jest.spyOn(Date, 'now').mockReturnValue(...) freezes time, and the issuedAt field becomes predictable.

Here the spy's main advantage surfaces - it can be put back. mockRestore() restores the real implementation: the wrapper comes off, Date.now shows the real time again. This works only with spies, because only a spy has something to restore - it remembers the original it wrapped. A jest.fn has nothing to restore: there is emptiness behind it. That is why mockRestore and spyOn are a pair.