According to the "unlocking" theory of phasehood, phases remain opaque until they enter into an Agreement relationship, after which their subconstituents become visible to further Agree operations (Rackowski & Richards 2005, et seq.). We provide novel evidence for this theory from a subextraction asymmetry in San Martín Peras Mixtec (SMPM): internal argument DPs are transparent for possessor subextraction, while external argument DPs are not. We argue that Agree-driven object shift, independently needed to derive VSO word order in the language (Massam, 2001) unlocks internal argument phases, thus feeding subextraction. In addition, we highlight two patterns in SMPM that further elucidate this theory of phasehood. First, embedded external arguments and nested possessors permit subextraction. While initially unexpected, we link this to conditions on phase obviation, building on Richards (1998). Second, wh-possessors move to the edge of all DPs, even those from which subextraction remains impossible. This demonstrates that movement to the phase edge is not sufficient for subextraction (though it may prove to be a necessary step, along with unlocking via Agree).