In order to guarantee the reproducibility of research results, large research communities, conferences and journals increasingly demand the provision of original research data. Since this is often not possible or desired, a certain tact and sensitivity is needed. With our method, combining provenance and evolution, we can identify the source tuples necessary for the reconstruction of a query result also in temporal databases. To avoid dirty data caused by the inverse evolution, we introduced the what-provenance, which remembers the data types of the source relation.