Although historians intuitively know what a corpus is, and exploiting a corpus is recognized as central in historiography, there has been little methodological focus on coming to a unified approach to the design and use of corpora. The massive expanse of digital information and processing capabilities over the past few years has also led to a diversity of approaches. After reviewing the history of the use of corpora in historiography, we examine how historians have taken possession of digital practice, and how it has interacted with the notions and uses of corpora: there are many diverse and somewhat incompatible perspectives. Next, we show how the corpus, as an input into historiography, can not exist anymore as an object and must be seen as a process or a pipeline. Then, its multiple and sometimes opposite perceptions can be unified, at the same time making history more scientific in the sense of Lucien Febvre's definition.