This talk will present an update on building a project on top of Wikidata, from our experience on Inventaire. Inventaire does two things regarding Wikidata: reuse its items data and extend it in a local database reusing a very similar data model. In both aspects, we regularly encounter difficulties as the community redefines some properties, or some P31 values, which may break the expectations on which external applications are built on. We would like to discuss advantages and drawbacks of considered work-arounds such as shadow items, data model bulk upgrade, different strategies to type and shape items.