The live Z scores are based upon a dynamic database and joint-time-frequency analysis, while the NeuroGuide postprocessed Z Scores are based upon a normative database of 1-minute averages. Thus they have different statistics, as is expected.
It is like taking one set of statistics of people's height, and then taking measurements of how far they can jump up and down. The static height statistics are more limited, and an unusually high or low value may be significant. When the signals are dynamic, moving up and down, it is more common for any value to be at the extremes.
As a result, live Z scores are generally about 1 to 1.5 standard deviations less than the static Z Scores.
This can be verified within NeuroGuide by using the JTFA viewing option, and watching Z Score computed using the JTFA. This will show Z Scores exactly the same as those that will come from the live Z Score DLL during training.
So there is a difference in Z Score range between live and retrospective Z Score, and it is expected from the statistics of the EEG. It is also possible to see the two types of Z Scores from within NeuroGuide, to see and verify this relationship.
Once this difference is understood, it is not a problem. It is still possible to set up Z score training protocols to train toward the mean (zero), or to bias any of the z score training parameters. We must simply understand that when we train, we are working with momentary statistics, not average statistics.