Conflicting size modifiers

[Zsar]

Example with "Large":

Table:Size in Beastiary\Monster Creation lists base attribute values as
18 8 14
for a size modifier of +8 -2 +4.

Size Quality in Advanced Race Guide\Core Races lists the size modifiers as
+2 -2 0

Example with "Small":

6 12 8 -> -4 +2 -2
vs.
0 0 0

Now, if they would at least have the same sum, but there they differ too with +10 vs. 0 / -4 vs. 0.

I had planned to use the Race Builder as a means to transform a static monster into a leveled race (ex.: "Sphinx" monster -> "Sphinx" race + 12 levels of "Sphinx" class now calculable as difference), but if the not-race modifiers are context sensitive...

Furthermore, if this context sensitivity is accepted, how would one find out the proper modifiers for the other size categories? "Tiny" is listed as well, with -8 +4 -2 vs. -2 +2 0, but that was it.
So, from smallest to largest:
-09 +8 -02 vs. ST DE CO
-09 +6 -02 vs. ST DE CO
-08 +4 -02 vs. -2 +2 00
-04 +2 -02 vs. 00 00 00
<< Medium >>
+08 -2 +04 vs. +2 -2 00
+16 -4 +08 vs. ST DE CO
+24 -4 +12 vs. ST DE CO
+32 -4 +16 vs. ST DE CO

I see no simple solution.
Though I would very much prefer size modifiers to be simply context independent.

... Before the Advances Race Guide came up, I had not been aware of this schism, as all races were either "Medium" - modifier free; or "Small" and there they have
-2 0 +2
and
-2 +2 0
so I simply assumed the racial attribute modifiers would be
+2 0 +4
and
+2 +2 +2
for a total of +8 (with the regular +2 CH) versus the +2 of the other choices.
Mayhap I should just stay with this assumption?

[Quandary]

i just don't believe they're meant to be compatable like that.
one reason why the bestiary guidelines aren't given as modifiers in the first place.

[Zsar]

So, a Svirfneblin's small size is different from that of a Pixie for no discernable reason?
... Well, this is how it is, obviously. Question: What to do about it?

• The "race" size modifier list must be completed.
  
• A well-defined way to assign creatures to either the monster or the race list must be described.

And I look for an elegant solution that either solves both points or renders one irrelevant.