# Reproducible benchmark

The repository benchmark is a deterministic workload over eight synthetic fixtures. It checks the full fixture path—evaluation, expected status comparison, boundary classification, and quality summary—rather than timing an isolated loop.

Run:

```powershell
moon run ./examples/returnplay_benchmark
```

Observed on 2026-08-21 in the local Windows workspace:

```text
samples=8
execution-count=24
repetitions=3
successful=24
failed=0
boundary-samples=12
success-rate=100
boundary-rate=50
status=pass;total=8;passed=8;failed=0
```

A single warm local invocation measured approximately 121.42 ms with PowerShell `Measure-Command`, including process startup and the Moon runner. This is an environment-specific smoke measurement, not a portable performance guarantee. For source scale, `scripts/source-scale.ps1` reported 5278 non-test MoonBit lines, 1253 test lines, and 6531 total `.mbt` lines; generated `.mbti`, build artifacts, documentation, and proposal files are excluded.