This project has one rule that has paid for itself more than any other: before building anything on top of a result, check that result against something independent. Twice now it has caught me being confidently wrong. The first time was a ragged edge on the thin skin of material that runs around the outside of every part. I had a tidy explanation ready, rewrote the whole thing to fix it, and it got worse. Measuring both versions side by side showed the old method had been fine all along, and that the raggedness I set out to fix had come from something else entirely: a grey fuzz along the edges that a different change had already cleaned up weeks earlier. I had been staring at an out of date picture and inventing a cause for it.
The second time I wanted diagonal braces inside a tall hollow arm, and I was sure that teaching the optimizer about buckling would produce them. It did not. It thickened the walls and left the middle empty. What actually worked was far simpler. The part was being optimized for one perfect load always pushing the same way, and real parts do not live like that. Add a second scenario where the load shoves sideways, ask for a design that survives both, and a brace appears immediately, as in the picture: same part, same material, one load on the left and two on the right. Obvious in hindsight, since a hollow box with two walls resists bending in its own plane beautifully and is helpless the moment you push it from the side.
Both times I had a story that sounded right, and both times it survived exactly as long as it took to measure the thing properly. About ten minutes. Intuition here is very good at producing plausible explanations and very bad at telling you which of them is true.