There are a fixed number of local representatives and the rest of the 120 available seats are made up by the total party vote. So if you won 10 seats but 25% of the party vote you would have 10 representatives and 20 list/party MP's.
It has also worked very well to give third parties more power and help keep the two main parties closer to the political center.
Marginalizing groups by splitting and packing is an affront to a representative democracy. This has become pathological now that there is software that can nudge boundaries by a single neighborhood block to chose the desired voter distribution.
But I think I know what you mean. But I also think that even if you don't split apart certain neighborhoods or regions that have very clear boundaries (which isn't universal), it's still very easy to go pretty far in the 5/4 direction or the 9/0 direction. So this standard doesn't really answer the question.
You'd almost want gerrymandering, but of a different sort. Traditional gerrymandering makes some districts win by 100/0 and the rest lose 49/51 to marginalize as many people as possible.
But perhaps more fair would be perfect gerrymandering - every district ends up with a 100/0 split. This means you get 5/4 at the top level, while every citizen is represented by someone they voted for.
Districts end up as more of an overlapping blur.
Neither, don't have single-member districts, use PR.
Optimize the variance of election outcomes. If pundits and such can't predict how the election will go, then and only then can we say voters really spoke.
Sure, algorithms can embed their own assumptions, but at least they'll be applied consistently based on known data. It's a "perfect is the enemy of good" kinda thing.
Say you have a 55/45 state, with 9 congressional districts. Roughly speaking, what is most fair? A 5/4 district split? Or, each district being 55-45, i.e. 9/0?
If I'm reading it right, it sounds like the efficiency gap would mean that 5/4 would be more fair, but that would actually require drawing districts in a way that pays attention to the political and demographic makeup of the geography.
I get that for many states, if the Democrats had the 55, then current gerrymandering means that they might only get 2/7, or two out of the nine candidates, which seems wildly off no matter how you look at it, but still, I have trouble identifying the principle of what fair actually is. And I don't like the answer of just arbitrarily picking a math formula that is used the same way everywhere. Because, in the absence of a principle behind it, it could be just as distorted. For instance, if 5/4 is more fair, then a geographically blind districting program could very easily end up with similar blended districts leading to a 9/0 state.