Yes, it will currently fail if the input has non-manifold edges or is non-orientable.
I’ll see if I can do anything to get it to heal some of these cases.
For now one option with bad meshes is to take a good mesh and wrap it onto the bad one.
The sphere projection technique I was trying here is a bit faster than starting from just the bounding box, though still takes ages to pull all the way in to the concave regions at this density. There are some internal variables for how far it allows a vertex to be projected, and I wonder if exposing these might allow it to reach a tight wrap in reasonable time.
Another option is to start from an isosurface. Here’s a go using Dendro to generate a starting mesh then TriRemesh to improve the triangle quality. It takes a while but gets there in the end.
triremesh_dendro.gh (10.8 KB)
(though whenever it is possible to fix non-manifold edges before remeshing, I think that will be fastest)