I built a repeatable way to check whether token-saving tools actually work

Hi, I’m Yu. Earlier this month I released Tura and wrote about why agent harnesses need long-horizon benchmarks, not just one-off demos. Over the past two weeks, I’ve seen more people publish tests of