At 12 minutes Jonathan explains why it happens and how to fix the flipped normals.
I did, however, when trying to reconstruct this from memory, connect the transform nodes a tad different, making one side change the location and the other side change both location and scale. In this way, I can skip the skip the "Geometry to Instance" node entirely and not have the normals affected.
My question is: is there a reason why one would be better than the other? Will I break the process down the road if I do it without that node? Or was this just a way Jonathan found to ensure we learned about this caveat?
