Execution-time sampling is great for latency stories, but it blurs the line between waiting and working.

CPU-time sampling gives you a cleaner picture of what actually burns cycles, which is the thing you pay for and the thing you can really optimize.

When two lenses disagree, you learn something about whether you’re chasing throughput or latency. That’s the conversation most teams need.