Performance¶

Run the reproducible orchestration-overhead measurement:

poetry run python examples/measure_overhead.py

It compares repeated direct no-op calls with complete one-round, one-iteration BenchmarkRunner.run() calls across multiple trials.

There is intentionally no universal pass/fail threshold. Scheduler activity, power management, virtualization, instrumentation, timer resolution, and interpreter builds materially affect microbenchmarks.

Performance changes should report:

• before and after commit identifiers;
• all environment information printed by the script;
• multiple-run medians; and
• confirmation that deterministic correctness tests remain unchanged.

BenchCore reports descriptive statistics. It does not claim statistical significance.