Job interviews are a good example of a kind of testing that we know doesn't work all that well but we keep around for lack of a better alternative. The others

Job interviews are a good example of a kind of testing that we know doesn't work all that well but we keep around for lack of a better alternative. The others are not good examples because they are useful in-and-of themselves. Giving good presentations and finding bugs are useful because they are useful, not because they predict something else. Pilot's licenses, like driver's licenses are based on tests where you more or less do the thing that you're being tested on. The test and the thing that it's supposed to measure are very similar and we can therefore use it to predict performance. In cases like that "teaching to the test" doesn't exist because practicing for the activity and for the test is the same thing. I'm not denying that educational testing is useful to quantify progress but such tests are often different enough from the underlying activity they're intended to measure that effort spent preparing for the test has little other purpose.