A "stub" is any test double that implements a so-called "stubbing", that is a preconfigured response to being invoked. A hand-rolled stub might be as simple as:

var fetchUser = function () { return {id: 42, name: 'Jane'} }

If fetchUser is then passed to a subject and invoked to facilitate some downstream behavior (like, say, printing the user's name), then that fake function could be called a stub.

This is an "unconditional" stubbing, however, because it will always return the same thing regardless of the arguments you pass it. Using unconditional stubs generally leads to insufficient tests, because the odds are that the real fetchUser invoked in production will behave differently if passed the wrong arguments.

As a result, most test double libraries are a good deal more sophisticated, handing users test doubles that can be configured to respond a certain way only when a particular stubbing is "satisfied" with the appropriate arguments. In testdouble.js, that would look like:

var fetchUser = td.function('fetchUser')
td.when(fetchUser(42)).thenReturn({id: 42, name: 'Jane'})

Now the subject will have to invoke the function fetchUser exactly as it should in production, because if it's called with different arguments, no stubbing will apply and undefined will be returned (which, presumably, would cause the test's assertion to fail).

For some more ideas on the sorts of configuration and features a stubbing API might enable, check out our documentation on stubbing in testdouble.js.