Contributing

Contributions are welcome, including from people who think the acute:chronic workload ratio is a bad metric. That view is well represented in the literature and this package is built to make it testable.

The most useful things

A bug report with a reproducible example. A failing test is better than a description. If a ratio comes out wrong for a particular load series, the series and the expected value are what is needed.

A hand-computed value the tests do not yet pin. Several tests check against numbers worked out by hand rather than against whatever the code produced first, because that is the only way to catch an off-by-one in a rolling window. More of those are always wanted.

A method from the literature that is missing, with the citation.

Setting up

git clone https://github.com/Lombergine/acwr
cd acwr
pip install -e ".[dev]"
pytest

Before opening a pull request

• pytest passes, including the doctests.
• ruff check . and ruff format --check . pass.
• mypy src passes.
• New behaviour has a test. New public functions have a numpydoc docstring with an example.

What will be pushed back on

A claim the package does not support. Nothing in the documentation should assert that the ratio predicts injury, or recommend a safe range. The thresholds that circulate in coaching material rest on the part of the literature that is in dispute, and this package deliberately does not ship them.

A silent fallback. Where the maths is undefined, such as dividing by a zero chronic load, the result stays inf or NaN. Filling it with a plausible number hides a real condition from the user.

A window computed on a sparse index. Every rolling computation runs over a continuous daily series. If a change makes it possible to compute a 28 day window over fewer than 28 days of index without the user asking for it, that is a bug even if the output looks reasonable.

Code of conduct

Be straightforward and assume good faith. Disagreement about the science is expected and welcome; personal hostility is not.