When running FBMN I notice some library hits are not very good (parameters: 4 shared peaks, 0.7 cosine score). On the library hits page, the cosine score meets the threshold but when I open the mirror plot the cosine score is different. For instance:

