Who says only supports can hold back the enemy team in order to ensure a safe retreat?
I like to play Jungle Nunu, and I can promise you that it happens very frequently that I have to stand still with my ult while my team retreats to ensure a bad engage doesn't end in an ace.
Every champion with some form of cc that puts themselves in danger and/or makes it difficult to escape at the same time can pull this off. And honestly, depending on positioning you can even force it with the threat of high damage burst.
It's not a matter of roles or even of champions. It's a matter of playstyles. And you cannot account for all playstyles because not all playstyles even currently are known. Strategies and playstyles pop up every now and then in a "Why did I never think of that"-way. And almost always it involves doing something that previously would have been considered a bad idea.
In the end it's a matter of personal philosophy: Do you prefer high accuracy or high precision?
Assuming that MMR aims to identify how much you contribute toy our team's chance of winning, Riot's system has undeniably the better accuracy, as this is literally what it's measuring. It's keeping track of which games you win and which games you lose, while also estimating how high your chance to win "should" have been.
However it fluctuates heavily. Given that you always either lose or win, but you never have a 100% estimated chance to do either of the two, your MMR will always change over time. That makes it impossible for Riot to ever give you a precise value for your MMR.
--> High accuracy, low precision.
Individual performance systems ultimately try to lower that fluctuation, by making the "measured" contribution less binary and allowing in extreme cases the gain of MMR despite losing a game. Eventually you could theoretically even reach a precise equilibrium for your MMR where every game, loss or win, your stats are equally good (given the circumstances).
But at the same time what you are measuring now is no longer the actual statistical contribution to your team's chance to win. You are measuring things that correlate with winning in some form. And while that could be accurate, there's no good way to tell whether that is the case, or whether it will stay this way.
--> High precision, low(?) accuracy.
Personally, I prefer Riot's method, as low precision fades with enough samples. That means that if the system has the wrong impression of your skill due to fluctuation, that might be bad, but at least you are capabale of "fixing" it yourself: Just play more games.
Low accuracy on the other hand, doesn't allow this. If the system in place is inaccurate for your personal playstyle then you're tough out of luck, and your best hope is writing a forum post about it and trying to get Riot to try and consider your playstyle in their formula.
