Although NIF 1.0 is specified the used ontologies can still change as they are not versioned as of now.
Should information already encoded in the URI be duplicated?
The URIs already contain all information to uniquely find the string and also calculate things such as begin and index and also the inclusion of str:anchorOf duplicates the reference string and might lead to scalability problems.
On the other hand many things are already duplicated and can even be simplified further and/or removed such as the class str:OffsetbasedString, the property str:anchorOf or the substring relations are useful for iterating the model, but are actually redundant.
Should the client or the server implement the conceptual interoperability?
For each tool the interoperability to one reference ontology has only to be implemented once, while on the other hand many clients would have to implement it. The difficulty is however that the servers have to provide exact and versioned implementations so the client does not get a conflict when merging.