Another idea it puts forwards is the use of structured data for labeling decisions, for instance, maybe you trust a providers' recommendations related to racism but not transphobia, or maybe you want the CSAM and DNI data, but not to block bridges. This is all possible if we've more structured data.