Interesting. This is in concordance with my own thoughts. The ideal space-filling "core" of a rete mirabile recuperator is (I argue) a checkerboard, with black

Interesting. This is in concordance with my own thoughts. The ideal space-filling "core" of a rete mirabile recuperator is (I argue) a checkerboard, with black and white squares flowing in the opposite direction.† This makes all adjacent channels opposite polarity (unlike a honeycomb) for maximum heat transfer, and also reduces internal drag and manufacturing complexity (unlike triangles). Mass manufacturing can be done via extrusion, or by gluing together 90° corrugated sheets. If the core consumes 80% of the total system volume this reduces 3D printing cost. Trees typically branch 1-to-3, which is probably optimal because evolution. For an 54x54 grid that means 1458 endpoints, or 6.6 branchings. That's one 2-to-1 branching (common on the first "trunk" branch of a tree) followed by six 3-to-1 branchings, for seven total branchings. This is a typical value for a real biological tree, so we're probably moving in the right direction re:biomimicry. In real biological trees the fluid flow slows as the branches get smaller, so it's probably good to emulate that. † arguably a honeycomb shell-and-tube is more efficient geometrically, but it's non-symmetric and (more important) the manufacturing and rigidity/robustness is questionable