Résumé : Filling the open archives remains the most pressing social engineering issue in scholarly communication. The poster shows the "who archives who" graph of a moderately large archive: HAL-SHS, an institutional national open archive focused on humanities and social sciences in France. In this graph, each dot is a person and each arrow shows which contributor archived which author, persons from the same lab having the same colour. This allows to see the various ways in which authors get their documents open-archived. Statistical analysis of the social network complemented with face-to-face ethnographic fieldwork, enables to define a few user archetypes: Agent, Power archiver, Solo, Contributed and Tester. We conclude that: 1/ Self archiving is about as important as proxy archiving. 2/ Contamination from user to user is almost nil (the graph is disconnected, lots of isolates, short paths, low reciprocity). 3/ Diffusion relies mostly on intradisciplinary personal affinity networks rather than institutional academic incitations or lab strategies.