This is about making type inference correct.
Correctness here means inferring types correctly and not producing false-positives.

This is the most important task and once it's done, we can provide a language server.