Isn't this an idealization myth of the scientific method? I'm pretty sure some kind of observation prior to forming the theory will account for many theories.

True that. It is like cat and mouse. You present a theory and validation and usefulness will come from experimental verification. And some theory comes from unexplainable observations. Michelson–Morley experiment and special theory of relativity will be taking the second path.

In a lot of situations, the observation is that the current favorite theory doesn't work for some observation, but the new theory explaining it only comes much later.

I disagree because we don't entertain theories which make no testable predictions. It seems logical that this rule makes observation the ultimate qualifier.