The problem I describe comes in the testing of the prediction. As you describe it, the prediction about rain making the roof wet can be directly observed, and thus (apparently) confirmed. However, not all predictions can be directly observed in this way. Take another example:
Observation: Marsupials are the dominant form of mammalian life in Australia, but relatively rare elsewhere in the world.
Hypothesis: Marsupials became isolated on Australia from other forms of mammalian life when the land masses drifted apart.
Prediction: Fossilized marsupials will be found on Antarctica.
The logical structure of this prediction and its confirmation is as follows:
If marsupials became isolated on Australia from other forms of mammalian life when the land masses drifted apart, then fossilized marsupials will be found on Antarctica.
Fossilized marsupials have been found on Antarctica.
This was trumpeted as confirmation of the hypothesis. However, because we cannot go back in time and directly observe the migration, we are forced to rely on affirming the consequence for the force of the argument.
Those who would sacrifice an essential liberty for a temporary security will lose both, and deserve neither. -- Benjamin Franklin
We see monsters where science shows us windmills. -- Phat