I suspect 1473 and -910 are scores of the same scorefunction. The fa_rep van Der Waals repulsion term is very sensitive to small changes in structure due to clashes. Look for the score breakdown by term and see if there’s a huge change in fa_rep?
It’s still quite possible that these are different scorefunctions; I know docking uses different scorefunctions for different parts of its trajectory (although I thought it reported the final score in Rosetta’s standard scorefunction, which the score app would also report). Both protocols should produce a score file (or a PDB with scores in it) – there should be a breakdown of the terms in the scorefunction (along with their weights). If the terms and weights present don’t match, you can run the score app with the docking scorefunction (the score app documentation ought to show you how but if not ask again).