A researcher cannot tell the future. Therefore any guess at a player's limit can be wrong (in both directions).
If there was some value (say 170) where the researcher would define a player's limit, that would be wrong because that implies the researcher can tell the future (cannot be greater than 170).
Therefore P(player becomes Messi | researcher makes guess) > 0 - i.e. it is possible, because you never know - that player might turn out to be the next Messi.
You don't need limits to ensure balance.
Sure, it becomes more difficult, but then again, why is that a customer's problem?
The quote you quoted was to do with this:By having a fixed and relatively consistent distribution, the game ensures a good spread of players of varying abilities, with only a few very very top players. That's the way it should be.
Bold bits added by me which I was referring to when I say "they aren't the same".Now, unless you disagree with that premise, changing the way PA works would simply replace one system - one that's really easy to code and manage because you just assign a single value to each player on creation - with another system that has the same result, but which requires incredibly detailed testing and maintenance. There's no point in simulating something in a complicated way of you can get the same results with a very simple method.
The overall distribution might be the same, but the mechanics might be different yet one might still be wrong.
Imagine if the underling FM model was that: Every player is guaranteed to reach their PA and PA is just generated from a random number distribution that exists in FM12 today. The distribution would be exactly the same as the supposedly-perfect model, but this new model is wrong, because it makes no sense to state that every player reaches their PA and PA is generated the same for every single player regardless of where they are.
I'm not sure the distribution of a PA-less model will be exactly the same but I don't see why having the same result as before should be a necessary goal. The distribution needs to be compared against real-life data to justify (or not) realism.