Consider this excerpt from this etymology of "lose":
Old English losian "be lost, perish," from los "destruction, loss," from Proto-Germanic * lausa- (cognates: Old Norse los "the breaking up of an army;" [...] ), from PIE root * leu- "to loosen, divide, cut apart, untie, separate" (cognates: Sanskrit lunati "cuts, cuts off," lavitram "sickle;" Greek lyein "to loosen, untie, slacken," lysus "a loosening;" Latin luere "to loose, release, atone for, expiate").
Among other things, this etymology claims that Sanskrit "lunati" and Proto-Germanic "* lausa" are cognate. Now, maybe it's the case that there are regular sound changes that could account for the fact that the words are phonetically different, but what bothers me is that "to cut off" is so distant from "to destroy, to lose" that I don't see why you would call these cognate at all, and any sound correspondences built partly on the assumption that they are would seem doubtful to me.
How semantically distant can words be and still be considered cognate, and what evidence is used to support the conclusion in such cases?