Latour bifurcates our traditions into the heterogeneous networks of domains (concrete traditions) and the homogeneous felicity conditions of modes (abstract traditions) and their values. Having spent a lot of time explaining to us how science, contrary to its image as an abstract tradition, is in fact a concrete impure assemblage of actors and trajectories, Latour is now explaining how nevertheless one can abstract out a pure value of objectivity that is served by this composite rhizomic domain.The further I'm getting through the book the more I'm liking it, however I am also more convinced of one of my initial impressions: that this is a work of pure philosophy. It is 'empiricist' only in the most abstract, philosophical sense (despite what Latour insinuates). Yes, Latour has done case studies on science, law, technology and so on, and to some extent aime is a synthesis of all is work to date. But he's written no case studies on politics, for example (Politics of Nature is another work of pure philosophy). The majority of the modes has no anthropological connections whatsoever - at least not yet.
Not that there's anything wrong with that as such but it is as though admitting doing pure philosophy isn't good enough for him and so it has to have these empiricist motifs tagged on here and there, clogging up the flow of concepts in order to make it all seem more real and grounded.
There is, of course, a very real sense in which his work is empiricist - that is, 'radical empiricist'. The whole endeavour is built on the foundation of James' flows of experience but this is a purely philosophical empiricism; ironically, it doesn't seem to flow all that directly from or into fieldwork.
However, as for this 'bifurcation between networks and modes' it's quite a weak bifurcation inasmuch as, I believe he states early on, every concrete situation is an event of both NET and one PRE or other. The limitless heterogeneity of NET is always combined with the particular essence of a mode. REF is not equal to science; science both exceeds it and REF exceeds science. So, there's still plenty of room for the heterogeneity and messiness of real practices.
I agree that his way of setting out the modes is conservative. However, every mode inserted into the schema will transform the whole assemblage. Perhaps it isn't the will to conserve and protect that is the problem but the absence of other modes that would, if recognised, transform the whole? That's my working hypothesis, I'm still working out what those modes might be.