There already exists several corpora that have been manually annotated in referring expressions and coreference chains. Nevertheless, none of them focuses on French language (or for annotations that are related to anaphora more than coreference). The DEMOCRAT project has produced such a corpus, with also a diachronic dimension. Its conception raised numerous difficulties, not only linguistic, but also in terms of the homogeneity of the annotations, as well as their verification and the evaluation of their quality. It is this dimension that we explore and discuss here, including concerns about annotation conventions and the evaluation of the annotations, a procedure involving the computation of the inter-annotators agreement. Thus, this article discusses the constitution and content of the DEMOCRAT corpus, in order to legitimise the exploitations that will be made of it.