Perf Benchmarker

It mandates 60 second runs with proper warmup, blocks parallel execution that pollutes results, and makes you re-run anomalies instead of cherry-picking good data. The tool expects structured JSON output between markers and won't let you take shortcuts with quick 5-second tests unless you're actively binary-searching for a threshold. It's opinionated about measurement rigor, which is exactly what you need when numbers actually matter for production decisions.