# Reproducible benchmark record

The benchmark executable uses fixed workload sizes and deterministic seeds. It prints numerical outputs and checksums only; wall-clock time is measured by the host shell so that compilation and execution timing are not mixed into the program output.

## Workloads

| Label | Workload |
| --- | --- |
| `gbm_terminal_paths` | 2,000 GBM paths, 64 time steps |
| `sobol_32d` | 4,096 points in 32 dimensions |
| `european_call` | 2,000-path European call estimate |
| `barrier_call` | 2,000-path down-and-out call estimate with 16 monitoring steps |
| `portfolio_scenario` | 64 positions evaluated over 256 deterministic shock scenarios |
| `implied_volatility` | Bounded bisection inversion of a Black–Scholes call price |

## Captured local run

Captured on 2026-08-23 with the repository's local Windows toolchain:

```text
moon 0.1.20260819
moonc v0.10.9+6e6c44045 (2026-08-19)
```

Command:

```powershell
moon run --target native --release benchmarks
```

Output:

```text
benchmark=gbm_terminal_paths paths=2000 steps=64 checksum=13791941313477298351
benchmark=sobol_32d points=4096 checksum=9200521295395616045
benchmark=european_call paths=2000 price=9.411281859280658 standard_error=0.32216128979573894
benchmark=barrier_call paths=2000 price=9.413602149695935 standard_error=0.31861990699007864
benchmark=portfolio_scenario positions=64 scenarios=256 checksum=2003668158890984229
benchmark=implied_volatility iterations=41 residual=3.766587042264291e-11 volatility=0.20000000000094442
```

One host-shell timing sample was collected with:

```powershell
$elapsed = Measure-Command { moon run --target native --release benchmarks | Out-Host }
$elapsed.TotalMilliseconds.ToString('F3')
```

Measured wall-clock time: **131.342 ms**.

This is a single local reference sample, not a cross-machine performance claim. Use the fixed output values for reproducibility checks and collect multiple timing samples when comparing hardware or toolchain changes.