It is not rare that a law obliges us to account for the execution of a penal measure. This article analyses the reports legally required in the context of conditional liberation measures. The conditional liberation measure lends itself readily to a double questio­ning, how and what are we to understand, because the successive laws (31 May 1888 and 5 March 1998) both presuppose the production of reports relating to the execution of this measure. Analysis of these different reports shows the importance of numerical data in understanding the execution of the law. This numerical data, while present in the two groups of reports studied, is not the same ; understanding the characteristics of the convicts or else their (good) management by the penal system leads to the production of different data. The competing goals of the judicial statistics provide a first tool to help us grasp this production of different data. The foucauldian distinction between «normation» and «normalization» power techniques also provides an approach to this problematic, although from a different angle.