Probably the greatest problem in AI right now is proof / benchmarking

I check every benchmark site weekly, compare harnesses and models, and its still pretty much just vibes