Reduplication patterns in Itelmen (Chukotko-Kamchatkan) present a prima facie challenge to the view that association in partial reduplication is always Edge-In (Yip 1988, McCarthy & Prince 1996). Closer investigation suggests that Itelmen reduplication may in fact be total copying, masked by an apocope rule. This solution is not obvious, however, as the apocope rule must be limited to ‘core’ or ‘native’ vocabulary. While Russian loans are reasonably transparent, the analysis requires making a distinction between cognates and loans from related Koryak, a distinction speakers are not consciously aware of. Positing that this distinction is part of the synchronic phonology provides a solution to other apparently unrelated phonological puzzles in the language. In addition to removing an apparent counter-example to universal Edge-In association, the proposals made here may also provide a small argument for the lexical stratification model of loanword phonology over a purely representational alternative.