# Reproducible benchmarks

The repository includes the CVRPLIB Augerat Set A instance `A-n32-k5` as a
small, redistributable regression fixture. Its published metadata is 31
customers, five vehicles, capacity 100, rounded Euclidean (`EUC_2D`) distance,
and a known optimum of 784. The exact source file is kept in this directory;
the MoonBit benchmark package embeds the same text and parses it through the
public TSPLIB parser.

Run the real solver benchmark with:

```text
moon run cmd/benchmark
```

The output reports the seed objective, best objective, gap to the published
known value, feasibility, route count, and iteration count. Results are
machine-dependent for elapsed time, so the repository does not claim a fixed
runtime or a fabricated quality number.

One checked run from this revision, using the default seed and 120 iterations,
returned seed objective `842`, best objective `827`, a `5.48469387755102%`
gap, a feasible solution, and five routes. Re-run the command above to measure
the result on the current toolchain and machine.

Source: [CVRPLIB CVRP instances](https://galgos.inf.puc-rio.br/cvrplib/index.php/en/instances/1)
and the Augerat Set A instance text distributed by the public CVRPLIB mirrors.