Probably the easier to understand (or longer) way of explaining the issue would go something like this:
During print start calibration, the printer tries to figure out at what height the nozzle is currently in relation to the bed as it doesn’t have this information from the previous print or any other sensor. For X and Y, that’s done with the end stop switches that are at the end of the rails and those can be heard clicking during the X and Y calibration process. However, there’s no end stop switch for Z. Instead, the end stop method is to probe the bed physically with the nozzle. Once it touches, that’s the zero and that defines the Z height that can be seen as “Result is z=” prints in the console. With a clean nozzle, this method produces good results. With a randomly dirty nozzle, the results are equally random. Because the Z offset is applied on top of this value instead of replacing it, the offset can’t be used to fix a bad result from the probe. It’s also not possible to automatically evaluate if that value was bad since there’s no other probing method available which could be used as comparison due to both pressure and eddy sensors relying on that same part, the nozzle.
I’m not sure if there’s any workaround solution existing for this to just manually set the Z height as there would still need to be some kind of Z end stop existing for the printer to know where the Z initially is. Probing with the pressure sensor using some other rigid part of the toolhead assembly could be a better approach but this hasn’t been planned for and there may not be any rigid enough part of the toolhead can make contact with the bed for making this work.