As I understand, its the opposite. String theory "predicts" any and every possible observation, then they cherry pick whatever predictions are already accepted as fact to say there is evidence for the theory.
The difference is that the string model doesn't (yet) have any observations which we could use to constrain it in an analogous manner. But in studying it, one would hope to find areas of the theory that indicate where those observations could be made.
The point being, it is not so simple to just look at string theory and say you can a priory determine it is flawed or wrongly motivated. That same kind of reasoning would likely have prevented Planck & Einstein from starting quantum theory.
I'm not sure what this means. Is it different from just making an observation? Eg, "this light in the sky is of this color at that location", or "this object drops at that speed".
No it doesn’t. There are a few good guiding principles that make it possible to build models that turn out to be an accurate description of nature. These guiding principles don’t manifest themselves in a set of axioms of a mathematical theory nevertheless they are there. And they are part of the theory, just not set in stone like the underlying mathematical structure. But that doesn’t matter at all as long as you are doing science and not math.
Take classical EM, measure the masses of the electron and atomic nuclei, and write down a suitable Hamiltonians. Viola, you have an explanation of almost every phenomena that’s relevant in our daily life and way beyond. Similar principles exist in elementary particle physics and lead to standard model.