And then you need to ask this question... should a character be viable no matter what abilities people choose?
No. No one should be asking that question. You can't make a viable DOTA team from 6 support heroes, you can't make a viable Overwatch team from 6 support heroes, and you shouldn't be able to make a viable MK11 character from 3 abilities that have absolutely no synergy with each other. That's just silly.
So, doesn't that sound like... uhm... preset variations? With the illusion that yes, you can choose, but if you wanna win, you will go for the best moves.
No, it doesn't sound exactly like preset variations. First of all, because people have other considerations than JUST winning. Even pros. And even for those pros who are EXCLUSIVELY concerned with winning, their own innate skillsets might favor sub-optimal characters or variations to such a degree, that it makes those characters or variations optimal
for them.
If you need hard proof, look no further than MKX. Here is
Combo Breaker 2018. Here is the most up to date
MKX tier list. Are Swarm Queen D'Vorah, Brood Mother D'Vorah, Tempest Kung Lao, Covert Ops Sonya, Flame Fist Liu Kang S tier? No. So why were they played? Well, because the pros prefer those characters/variations, or because the pros are actually better with those suboptimal characters/variations than with higher tier ones.
And if that all isn't enough to make the "one variation used per character" argument sound bunk...
Let's try a thought experiment:
If a game launches with only two characters with one "variation" each, and we give them both a random power level between 0.0 and 10.0, how long will it take the community to figure out which character is stronger? Chances are extremely high that the difference between them will be more than 1.0, i.e. extremely noticeable. Probably wouldn't even take a day, right?
Now let's make it 25 characters with one variation each. How long to rank them all with no mistakes? Suddenly instead of one character being 9.4 and the other character being 3.7, you have 25 characters all over the spectrum. One might be 9.6, another 9.3, another 9.1. Telling the difference between them could take weeks or even months.
Now let's make it 25 characters with 3 variations each. That's 75 unique character+variations total. One might be 9.9, another 9.8, another 9.7, another 9.6, another 9.5, another 9.4. How long to rank each variation with no mistakes? Well, we know from MKX, that in many cases it did take months for a character to "turn out" to be a lot stronger than people initially thought.
Now let's make it 25 characters with 10 abilities each, for a total of 720 possible variations each. That's
18,000 unique character+variations total. How long to rank each variation with no mistakes? One decimal isn't going to be enough to distinguish between them anymore, we're going to need two or three.
Obvious conclusion is obvious. The more options you give players, the more "good enough" possibilities you create, the longer it takes them to even figure out what the "good enough" possibilities
are.