Things I’ve thought about and want to write about.

1. Why scalable representations matter for scientific discovery
2. Safe pieces can still compose into unsafe systems
3. When is a pipeline a scientific contribution?