I agree we need to move to a more continuous approach. We should constantly update a (digital) model of our understanding of a topic, and each (attempted) update should reevaluate against all existing cases.

Current papers are too fragmented, and primarily accumulate knowledge in those humans that actively participate in the field. By them reading the papers, evaluating and combining them. To transfer that knowledge to someone new is a process measuring in multiple years (ie a PhD). And when a person then ultimately exits their field, a lot of the accumulated understanding is lost.