The main approaches to knowledge representation and sharing (KS) focus on easing the exploitation and exchange of knowledge representations (KRs) between particular agents or for particular applications. These approaches lead to the proliferation of mainly independently developed KR bases (KBs) which are mutually partially redundant and contradictory, thus restricting a more general KS. This article shows how a shared KB edition protocol can support the incremental and cooperative integration of fragmented knowledge into a (networked or not) consistent well-organized KB, without manual or automatic selection between contradictory KRs. To that end, the protocol helps and enforces i) the setting of particular relations between “competing” KRs (i.e. those that are contradictory or at least partially redundant) to compare them and justify their joint existences in the KB, and ii) a representation of them that keeps the KB consistent. Besides providing more knowledge search and inference possibilities, these relations can be exploited by default or user-defined rules for KR filtering and choices between them. This article also shows that this approach can be implemented in various ways and be used with most inference engines.