Strongly agree- this is how I “evaluated” languages pre-agents. though I suspect this would bias results in favor of whatever has best signal to noise for boilerplate from stackoverflow/reddit , rather than what LLM’s “””reason””” best with. (Presuming those aren’t quite one-and-the-same)