# HRVKit Benchmarks

This file records a reproducible local integration baseline for the deterministic
benchmark workload. It is an observation, not a cross-machine performance
guarantee.

## Workload

The CLI runs the complete cleaning, time-domain, frequency-domain, nonlinear,
and report-feature pipeline over the built-in 256-sample resting fixture, with
10 benchmark repetitions:

```text
samples,repetitions,feature_count,mean_rr,rmssd,total_power,quality_ratio
256,10,35,799.8931407352728,2.368779541044154,14633.821047418522,1
```

## Local measurement

Measured on 2026-08-19 in the Windows development workspace with
`moonc v0.10.7+bc794d341`. After two warm-up invocations, six consecutive
`moon run --target native --release` invocations took:

| Measurement | Value |
| --- | ---: |
| Samples (ms) | 408.96, 498.91, 462.20, 409.36, 401.49, 415.82 |
| Average (ms) | 432.79 |
| Median (ms) | 412.59 |
| Tests | 265 passed on wasm, wasm-gc, JS, and native |
| Native coverage | 5,233 / 7,297 lines (71.71%) |
| Production MoonBit source | 20,006 physical lines |
| Non-comment, non-blank production code | 17,059 lines |

Commands:

```bash
moon test --target native --deny-warn --enable-coverage
moon coverage report -f summary
moon run --target native --release cmd/main -- --action benchmark --format csv --repetitions 10 --sample-rate 4
```

Production source scale excludes `_test.mbt`, `_wbtest.mbt`, `_build`, and
`.mooncakes`. CI recomputes the count rather than trusting this document.