# Benchmarks

The checked-in `benchmarks` package uses deterministic fixtures with three
workload sizes. It reports the number of parser/index iterations and the
resulting module, edge, and diagnostic counts. It does not use random input.

## Local run

```bash
moon run benchmarks --target wasm-gc
```

Observed on 2026-08-22 in the shared Windows workspace with MoonBit
`0.1.20260814` / `moonc v0.10.8+8606a5800`:

```text
small: iterations=100, input_lines=4, modules_seen=300, edges_seen=400, diagnostics_seen=0
medium: iterations=50, input_lines=13, modules_seen=600, edges_seen=800, diagnostics_seen=450
large: iterations=20, input_lines=37, modules_seen=720, edges_seen=960, diagnostics_seen=660
```

PowerShell `Measure-Command` around the same WasmGC command produced three
wall-clock samples of 355.76 ms, 122.12 ms, and 135.29 ms (average 204.39 ms).
These are local process measurements, including launcher overhead, and are not
portable performance guarantees.