Yes I am a huge believer in the algebraic view.

That's true even for implementations; I want to see the general ideas implemented as libraries so the implementation isn't an monolith mashing up ideas from various of many papers, but a composition of libraries.

Yes, there is a lot of rich stuff in the MLSub thesis which is hard to follow, but the functional perl approach here I also find confusing so there is no attempt to distill the essence of what's going on.

Towards the bottom of the accompanying paper (my PDF viewer lags less than my browser!) there is a little bit of theory discussion, but I would like to see more. It would be very good if someone could take some of the tricks here and present them more abstractly.