As far as I can tell, most coherentists want to match theories with reality too, because truth doesn't really have any other useful definition. The goal is not to be coherent within a random and reality-detached set of sentences: the goal is to be coherent with the whole of science. When a scientists rejects (assigns very low probability to) the observation of a perpetuum mobile on the basis that it contradicts the laws of physics, that is a standard coherentist move. This is another one. The goal is to avoid having to waste time and costs on non-fruitful data gathering. Ultimately the only thing that is rejected is that blind data-only approach that may be considered the straw-manning of the correspondenceist position, except that one is actually unfortunately used too much. A coherentist will simply not spend money buying an airplane ticket to check if someone's garage has a dragon, the proposition contradicts so much we already know that the very low prior probability does not worth the cost. You may as well call this a wiser version of correspondencism, the barriers are not exactly black and white here. This is unfortunately philosophy, so fairly muddy :)