In a (non-linear) fit, and are strongly
correlated
so that is approximately preserved.
This integral sums all measurements, and is very well determined
experimentally, and so defines best determined mix of theoretical
parameters.
Unfortunately, we only want (predicted by theory) and
can not use fact that a particular linear combination of
and is predicted much better than either parameter on its own.
and 
are strongly
correlated
is approximately preserved.
(predicted by theory) and
can not use fact that a particular linear combination of 
and 
is predicted much better than either parameter on its own.