>The new theory cannot just beat the old theory on the new data, it has to beat the old theory on the old data,

Erm, yes - if you come up with a new theory of gravity, it had better explain the current orbits of the planets.

Suppose I have a wonderful new theory of gravity, it doesn't work for anything on earth, or anything you can measure in space - but that's ok because that's old data and doesn't count.