From what I understand, FreeCAD is limited by the OpenCascade kernel. I wonder if brlcad could help to address this issue in any way.

It's limited more by the way it uses it.

No, OpenCascade just kinda sucks to use.

In what ways is freecad using opencascade incorrectly?

brl-cad is CSG modeler with some BRep support. Opencascade is a BRep modeler that supports boolean operations. The industry gave up on CSG and went BRep decades ago.

openSCAD is also a CSG modeller, and quite popular in 3d printing.

Not entirely true. As an explicit edit method, sure, but several of the big-name non-parametric (i.e., solid modeling) commercial CAD systems are fundamentally still CSG representation under the hood. They just hide it well, typically under the guise of feature edit nomenclatures or editing constraints.

Can find indications it's CSG under the hood if it won't let you directly modify a surface without selecting some "convert to editable representation" option. TinkerCAD is an extreme example that's basically a pure CSG modeler (and probably the most popular CAD system to date, but I digress), but you won't find the term union, intersection, or CSG anywhere in its GUI. Solidworks and NX do a really good job hiding it.

I guess we have a disconnect. I am talking about modeling kernel geometric definitions and data structures. Whether NX has or doesn't have the boolean terms (union, intersect ...etc) doesn't change the fact that parasolid is BRep. Just because parasolid supports boolean operations, doesn't mean it is CSG.

IMHO: Any shape defined by a CSG modeling kernel can be defined in a BRep modeling kernel. The reverse is not true, short of some kind of 'hack'. BOT(bag of triangles)? Hybrid with mesh modeling?

The topic has come up and been a point of discussion, especially developing something akin to the ACIS or Parasolid kernels, but open source. In many areas, BRL-CAD's capabilities far exceed what OpenCascade provides, particularly w.r.t. CAE features, and in others it is noticeably lacking (e.g. export to polygonal). Couple years of effort have been invested in developing a new C++ API for exactly this purpose (codename "MOOSE").

Current dev focus is on conversion, rendering, and GUI modernization.

I wonder why do you care about rendering? There are many good open source renderers out there, now. For instance can't you just export the geometry and render with Yafaray?