I was taught that science needs to be predictive, not just descriptive. It's easy to look at something and guess how it got there and why it behaves the way it does, but to be legitimate science your theory needs to predict unobserved phenomenon. If you find those phenomenon when you go looking for them, then that is support for your theory.
As written, this sounds overly restrictive, but I don't think you mean that. You say "to be legitimate science your theory needs to predict unobserved phenomenon." Yeah, but that does not mean that you cannot develop your theory based on
observed phenomena. As
this site explains:
The first question any theorist must ask is whether this theory can account for the masses of data that have been taken through history... Prediction of the past is the vital first step of science.... The next step asks whether your new theory continues to explain the world as new data comes in....
Obviously, as new data comes in --
that's the so-far "unobserved phenomenon" that you mention. Newton observed
past motions due to gravity and somehow figured out an extremely good estimate that
described those motions. That estimate also
predicted how
future motions due to gravity would take place.
AFAIK, it's only Einstein whose theories predicted phenomena that no one had ever observed before. Warped spacetime? Gravitational waves? (Only recently observed!) Gravitational lenses? Time dilation of distant receding objects? Well, the precession of Mercury's orbit
was observed, but no one could explain it... until Einstein.
Normally though, scientists first observe phenomena, then develop a theory, not simply to describe the phenomenon, but to explain how it
and any future such observations play out. Any theory really has to be able to be
tested. If the theory doesn't pass the test, it's got to be discarded. An interesting quote regarding quantum theory:
"...quantum theory boldly exposes itself to potential falsification on a thousand different fronts. Its record is impressive: quantum theory passes every test we can devise. After sixty years of play, this theory is still batting a thousand."