# Benchmark evidence

This page records a reproducible local observation for the native release
benchmark suite. The values are not a cross-machine performance guarantee.

## Environment

- Date: 2026-08-24
- OS: Windows 11 / Windows version `10.0.26200.8655`
- CPU environment: Intel64 Family 6 Model 186 Stepping 2, 16 logical
  processors (the sandbox does not expose the marketing CPU model)
- PowerShell: `7.6.4`
- Moon CLI: `0.1.20260819 (fc2a4ee 2026-08-19)`
- Moon compiler: `moonc v0.10.9+6e6c44045`

The toolchain was verified with:

```text
moon version --all
moon 0.1.20260819 (fc2a4ee 2026-08-19) ~\.moon\bin\moon.exe
moonc v0.10.9+6e6c44045 (2026-08-19) ~\.moon\bin\moonc.exe
moonrun 0.1.20260819 (fc2a4ee 2026-08-19) ~\.moon\bin\moonrun.exe
```

## Command and result

Command:

```text
moon bench lib/pipeline --target native --release
```

Observed output:

```text
name                  time (mean ± σ)         range (min … max)
invoice_normalization    9.37 µs ± 217.69 ns     9.04 µs …   9.64 µs  in 10 ×  11219 runs
document_set_validation   33.19 µs ±   1.80 µs    31.69 µs …  36.54 µs  in 10 ×   3173 runs
diagnostic_report_json   36.48 µs ±   2.77 µs    31.86 µs …  39.50 µs  in 10 ×   3085 runs
```

The benchmark source is
[`lib/pipeline/pipeline_benchmark_test.mbt`](../lib/pipeline/pipeline_benchmark_test.mbt).
It measures invoice normalization, complete document-set validation, and
diagnostic JSON report generation over a fixed invoice/PO fixture.