Sure, but what if the upscaling algorithm misinterpreted a P as an F? Without manual supervision/tagging, there's an inherent risk that this information will ha

Sure, but what if the upscaling algorithm misinterpreted a P as an F? Without manual supervision/tagging, there's an inherent risk that this information will have an adverse effect on future models.