Why are you conflating most individuals not being average with determining the average being impossible? Are bears on average stronger than humans are humans on average smarter than chimps? You can answer these questions just fine, despite most bears humans and chimps not being average.
I'm not.
What I'm saying is,
you're using these averages, aren't you? Which means they're coming from data about those creatures, within the simulation. That's what simulationism is all about. But those averages are
deeply misleading, because they give the impression that EVERY member of the population fits those averages. And those averages are, quite literally,
the only thing that the racial ability score bonuses can be derived from.
But if you're deriving bonuses
that every member of the race must have, you are enforcing that
all members of the race are average, physiologically. That is factually untrue, and specifically a purely-gamist abstraction applied to real populations which
will deviate from that, sometimes very significantly.
Yes, training should matter. But we're not talking about training, here. We're not talking about ASIs from levels. We're talking about
racial ability scores. And it is simply, factually,
not true that "the average dragonborn is stronger than the average human" equates to "
all dragonborn are naturally +2 Str compared to
all humans, regardless of other factors." Those modifiers simply, factually,
cannot capture the real variability that real, living populations express. No simulation predicated on this abstraction will produce results that conform to an expectation of real-population-like dynamics, because that's simply not how real populations
work.
The average, as a numerical value, exists. That does not, in
any way, imply that that reflects a fundamental attribute of the population in question. Deviations--sometimes dramatic ones!--
will exist. Those deviations are far more likely to occur in special subsets of the population that differ from the norm in other ways, and adventurers are about as unique a subset as one can get.
Hence: The way the
actual averages of
real populations work exactly contradicts this effort at simulating things. An
actual simulation of things would indicate that, while the average exists numerically and describes a
trend observable in that population, it not only
does not but
cannot be used to preclude the existence of (say) a few elves that are just as uncoordinated and slow as an uncoordinated and slow dwarf. (Or, likewise, the existence of a few dwarves that are just as dextrous as the most dextrous of elves.) And, thus, we get a step closer to
actually simulating a population of distinct beings by having not just variability, but varying degrees of variability, dancing around those central tendencies, which will never be directly observable by players because they aren't taking surveys of 4000 elves to find out what the average elf's dexterity is to begin with.