Rosetta scores scale by the size of the protein. Larger proteins tend to have more negative scores. Furthermore, how well a protein scores can depend on how extensively you’ve optimized it. (More aggressive optimization means that Rosetta has a chance to find better scores, so you’ll get overall lower scores with more agressive optimization options, even for nominally the same structure.)
So absolute total scores are typically not something to be too concerned about. As long as they’re roughly better than -1 unit per residue there’s nothing particularly concerning about any particular absolute score. (e.g. if a 200 residue protein is only scoring -123 units, then there might be issues you want to dig into further, but if it’s scoring -300, I wouldn’t be concerned from the magnitude of the score itself.)
Instead, you typically are more concerned about relative scores between different models, and how the scores of two different structures compare to each other.
The scores of single structures can point out issues, but that’s typically not the total score which is pointing that out, but rather the per residue scores. Sometimes you’ll have a protein which is doing decently well, but there’s a particular subregion which has issues. The total score washes all that out, but the per-residue scores will show decent results for most residues, but a drastically lower score for particular residues or regions of the protein.
.
