Analysing situated interactions requires contextual information of the learning situation to be present and connected to the collected data. It is essential if we want to construct a research object usable by various research teams and disciplines, which could lead to replication of experiments and induce cumulative or contrastive analysis. At present interaction data are generally not contextualised, not exhaustive or simply not freely accessible. We define a learning and teaching corpus by the description of its required components: research protocol, learning design, structured interaction data (interactions, productions, and log files), rights and informed consents, analysis. The resulting structure makes this corpus usable by the research community and permits the capitalisation of analyses. This proposition is illustrated by the construction of a corpus based on an authentic learning situation named “Simuligne”. This work is then discussed according different viewpoints: ethic and rights, standardisation efforts in the distance learning community and the new trends on tracks analysis tools interoperability.