There's some pure and delightful about doing performance optimization on research code. Probably no one else will ever run it, so I'm just optimizing for me. Feels less stressful than trying to decide whether it's "worth it" to optimize code that's used by someone else.