I've been reading John Mearsheimer's essay 'The False Promise of International Institutions' and am somewhat irked by its many weaknesses. So much so that I don't really know where to begin. (Some of the contradictions he makes are so blatant as to be funny.) I think the most objectionable aspect is the role he assigns to 'theory.'
In my opinion: Any given international incident is sufficiently different from all other comparable incidents that responses to it will always be conditioned, primarily, by situational rather than theoretical knowledge. Theory will never determine practice. However, without theory in the broadest sense it is impossible to make sense of any situational details. Therefore, theory can never determine practice but practice depends upon theory nevertheless.
'Realism' (in the vein of Mearsheimer, not generally) is a bad theory because it aspires to a level of power that could determine the response to any given incident. As such it abstracts from almost every situational detail. As such it is flawed, perhaps even dangerous. It ignores the vast majority of what goes into any given situation and, what's more, is proud of this fact. And, of course, it fails.
The complexities of international politics demand empiricism with theoretical support; the half-baked social physics of latter-day realism is less than useless. By portraying all action that conforms to its assumptions as necessary action, action that it would be irresponsible not to take, then it justifies - a priori - that action.
That said, more subtle varieties of political realism exist and are both deeply necessary for political theory and largely neglected by mainstream IR theory.
Political realism must not determine any given view of any given incident but it must not be eliminated from it either. It is both too limited and too important for the role that realists ascribe it. It can never fulfill the role they dream of for it and the more they try to shoe-horn it into that role the less relevant it becomes for its proper task: framing the situational specificities of particular incidents.
Totally empirically unfounded transhistorical, transcultural (so, really, universal) forces of 'human nature' are just too unwieldy and monolithic to, effectively, automate intellectual and political responses to intensely detailed particularities, as Mearsheimer apparently wishes they could.
Political realism, at its best, is a theory of contingency and the lack of assuredness or foundation when encountering concrete particularity. It is itself an argument against grand theory of this sort.
But this is not the 'realism' of the 'rich tradition of realism' that Mearsheimer claims, without citing any evidence whatsoever, goes back 700 or 1200 years (he claims both).
Grumpy proclamations of the importance of 'external reality' and theories of correspondence thereto are placed in quite comedic contrast when grandiose assertions about the millennia long endurance of a single theory are provided without even a hint of evidence for such an improbable and unbelievable trajectory.
And so it is for so many who defend epistemological realism and 'science' in IR: when they take on critics of their position they fail to correspond to those critics' arguments in any way!
Realism of all kinds deserves better. So does science for that matter.