@danderson Yes, the App Link changes that Assembly3 needs to work were merged into upstream #FreeCAD. Sadly, because Assembly3 depends on a forked version of the SolveSpace solver that is licensed under the GPL, it won't be integrated into FreeCAD proper (unless, I suppose, someone writes a replacement constraint solver that can do the same work but is under a license that FreeCAD will accept).
There has been a lot more than just Assembly3 in the #Realthunder fork, though. In particular, his topological naming stabilization work seems to me to be as robust as SolidWorks was when I last used it about a year ago; that is, it is of course possible to create topological nonsense by large changes to the upstream tree, but I'm much more likely to be comfortable not creating a profusion of datum planes in a part with his fixes. That work is planned to be one of the major headlines of work for the next stable release of FreeCAD, currently in development.
The downside of his fork right now is that because he's been so focused on getting his topological naming improvements merged, he hasn't yet merged released FreeCAD 0.20 into his branch, so it is a little behind on some of the stable release work.
If you haven't looked at it, Assembly4 (assembly without constraints), which like Assembly3 depends on the App Link work, is also interesting. Driven by relationships between datums across parts, it doesn't run into the problem of misinterpreted constraints because it doesn't use constraints, and it makes animating assemblies natural.
I expect it will generally be strictly faster to compute an assembly than to solve constraints; for at least some assemblies, it will be much more convenient to express constraints than to design for assembly by matching datums.