What?

When the docs say you can't rely on something, it means that the behaviour isn't part of the public interface. It's liable to change without notice.

In fact, docs are more likely to say that about predictable things, since those are the very behaviours which users are likely to assume are reliable.

This sort of thing is more of a problem when you do too much learning through experimenting. You can't tell whether things you learn via experimentation are defined behaviour or an accidental side-effect of implementation, so you need to check them against the docs before relying on them.