# Benchmark results

This document records a real local run of the benchmark package. It is an
end-to-end command measurement, including the cached `moon run` launcher and
the benchmark process; it is not presented as a microbenchmark of one method.

Command:

```powershell
.\benchmarks\run.ps1
```

Workload output from the same run:

```text
iterations=100000
checksum=19620206.497234184
```

Observed local Windows JS-target timings for five runs were 160.70 ms,
150.36 ms, 174.50 ms, 171.13 ms, and 191.45 ms (mean 169.63 ms). These
numbers include the cached `moon run` launcher and are provided as a
reproducible baseline for future changes, not as a cross-machine guarantee.

The checksum is deterministic for the fixed workload. Timing is intentionally
machine-specific and should be regenerated on the target runner rather than
treated as a portable performance promise. The Windows recording uses the JS
target because the local stable-toolchain native runtime fails before project
code is executed when the available MinGW headers do not declare `rand_s`.