Records Disposition in Japan’s Central Government: Norm, Practice, and the Applicability of AI

Abstract Public records management in Japan’s central government is structured around a normative framework, yet how this framework is implemented at scale has rarely been studied. From the Administrative Document File Management Register, published by each central government agency on the e-Gov portal under Article 7 of the Public Records and Archives Management Act, we analyze approximately 111,000 records across sixteen agencies whose Register we can retrieve near-completely. We obtain four principal findings. (1) Transfer rates to the National Archives fall short of full transfer in every standardized category the Guidelines designate for full transfer: about one in three records expected to be transferred are instead recorded for disposal. (2) Records created after the 2011 Public Records and Archives Management Act are recorded for transfer some twenty-seven points less often than earlier ones. The fall is a change in which kinds of record the Register holds rather than in how any one kind is treated. (3) The category a record is filed under says far more about its fate than the agency or ministry holding it, even though each agency schedules its own records. (4) A machine-learning model reproduces the recorded disposition closely enough to narrow the records that warrant individual inspection to an identifiable minority. Together, the findings show a systematic gap between norm and practice under delegation, and indicate that AI-assisted review can bring departures from the norm to attention at scale.