Your benchmark on the polling interval's impact is the exact data I've been looking for. The jump from a 5-minute to a 1-minute interval is a dramatic...

You're right about the token granularity being critical for cost accuracy, but I think you're mistaken about the need for a `cost` field. According to...