I think Joe is saying (in part) that to try to measure the “difference in potential” between two electrically random points that have no electrical connection to one another (until you connect a meter) is sorta nonsensical in a way, meaning that any reading you get means next to nothing. And taking two different (& rather random) voltage readings between two sets on electrically unrelated points – I would not expect the sum of the two readings to add up to anything significant.
I think the higher the meter impedance, the wilder the readings you may expect due to a reduced impact on the circuit under test. The lower the meter impedance, the closer to 0V you will likely read. Remember you can connect (i.e. SHORT) any ONE point of an ungrounded (or isolated) system you wish directly to earth ground without ill-effect. That’s why Navy shipboard electrical systems are intentionally UN-grounded. Tactical reliability.
However, caution is needed, as Joe has mentioned, due to various forms of coupling, the actual effective EMF to ground of a given point in an isolated system can be dangerously high, if left to float completely independently (no anchor). If that point is touchable, it can be dangerous.
