The utilities displayed in the summary are rescaled to "zero-centered Diffs". The utilities are zero-centered within attributes for tabulation to you, and the size of the utilities is expanded using a multiplier such that the average difference between best and worst levels within an attribute (within a respondent) is 100. You can see a description of that rescaling at:
We do this rescaling so that each respondent gets approximately the same weight in the averages. Otherwise, some respondents (who have larger magnitude of scaling in ACA, due to their answers in the Calibration Concepts) will get too much weight.