That sounds... contradictory, isn't it? Python is immensely hostile to static analysis (note that I write this as someone who writes Python for a living, using

That sounds... contradictory, isn't it? Python is immensely hostile to static analysis (note that I write this as someone who writes Python for a living, using pyright/mypy). If you remove the ability to perform static analysis from Lean, what's left? Or do you mean a Python-like syntax on top of Lean4? Note: Pretty sure that libtorch is written in C++ and CUDA.