God that reasoning is so broken. Not that I am getting at you.

How about viewing a swing in votes as a change in probability, and hence the natural (naive) model is to multiply each pior vote share by the ratio that the vote swing represents. As all probabilities are just ratios, after all.
This never yields zero, so no counterintuitive conclusions...

It's wrong when the probability of change is by definition not distributed across all voters. say if you knew that all pensioners are voting green next election, or all Gen Z are were voting Reform. These are deliberately extreme cases but illustrate where there isn't extreme factionalism, it's reasonable to consider shifts in probability, IMHO.

Though it has to be said thst I
do worry about factionalism and it's odd outcomes.

Finally, when not uniform, the ratio changes by area, as you've alluded to for some obvious constituencies.