Pharos is a collaborative application which enables users to share document annotations. Annotations of a same subject are stored altogether in channels, and channels are replicated to improve performance. What characterises data replication in a collaborative application like Pharos is the way users see data. If, for instance, mutual consistency is required, then collaboration should be synchronous. On the other hand, if remote users can work in disconnected mode, collaboration can be asynchronous with weaker constraints on data freshness. In this paper, we focus on asynchronous replication which is typically required on the Web. We propose an optimistic replication model based on lazy group replication and a protocol to detect and resolve potential conflicts to refresh the replicas. This protocol is based on the ordering of write operations at each site using its timestamps values. Careful log management is the key to its implementation. We describe the implementation of our model i...