Compared to spatial pattern recognition which is what the ego cells are currently doing, sequential pattern recognition presents itself as another essential building block.
Initially, I thought that some kind of memory was going to be needed to keep track of scopes (i.e. code block content vs paragraph content). Then, I thought maybe just spatial pattern recognition would be needed to identify scopes. But after realizing that links and inline-type blocks can wrap to new lines, nearest-neighboring rules can no longer solve the problem. This makes sense though. My original assumption about parsing markdown was that only top-down pattern recognition would be needed — my own sequential reading of the text went unnoticed by myself.
This leaves me with either using a memory and/or using some kind of sequential pattern recognition cell type. I'm still not sure about how, but it's clear that since the nature of written language i.e. markdown is sequential, a cell for this would make sense to exist.