Strictly speaking, the only quantity that has any real meaning is the difference between two RAW utility scores within the same attribute.
So if you have raw utility scores and you have one level with utility 2 and one with utility -2, the difference of 4 is meaningful.
If a different respondent had a corresponding utility difference of only 3 between the same levels, then the second respondent you could judge to have less modeled difference between those two utility scores.
The exact number, e.g. 2, is meaningless, but differences matter, and they are only strictly comparable on the raw scale or uniformly applied functions of the raw scores (e.g. a logistic transformation to turn utilities into probabilities like we do in the simulator).